<?xml version="1.0" encoding="utf-8"?><feed xmlns="http://www.w3.org/2005/Atom" ><generator uri="https://jekyllrb.com/" version="3.10.0">Jekyll</generator><link href="https://martineliseo.es/feed.xml" rel="self" type="application/atom+xml" /><link href="https://martineliseo.es/" rel="alternate" type="text/html" /><updated>2026-11-16T08:00:54+00:00</updated><id>https://martineliseo.es/feed.xml</id><title type="html">Falsifyer</title><subtitle>Cyber threat intelligence from open sources: adversary profiles, tradecraft, and the analysis behind them.</subtitle><author><name>Martín Eliseo</name></author><entry><title type="html">Grandoreiro: a banking trojan that survived its operators</title><link href="https://martineliseo.es/grandoreiro-actor-profile/" rel="alternate" type="text/html" title="Grandoreiro: a banking trojan that survived its operators" /><published>2026-10-29T00:00:00+00:00</published><updated>2026-10-29T00:00:00+00:00</updated><id>https://martineliseo.es/grandoreiro-actor-profile</id><content type="html" xml:base="https://martineliseo.es/grandoreiro-actor-profile/"><![CDATA[<table>
  <thead>
    <tr>
      <th> </th>
      <th> </th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td><strong>Product</strong></td>
      <td>Malware profile</td>
    </tr>
    <tr>
      <td><strong>Handling</strong></td>
      <td>TLP:CLEAR — open sources only</td>
    </tr>
    <tr>
      <td><strong>Version</strong></td>
      <td>1.0.1 (2026-10-29)</td>
    </tr>
    <tr>
      <td><strong>Review trigger</strong></td>
      <td>A conviction, a new distributor, or a vendor resolving the operator model</td>
    </tr>
    <tr>
      <td><strong>Next review</strong></td>
      <td>2027-04-30</td>
    </tr>
    <tr>
      <td><strong>Author</strong></td>
      <td>Martín Eliseo</td>
    </tr>
  </tbody>
</table>

<p><em>Every claim below carries a source and is graded on two independent axes. Where vendors
disagree, I say so — and on one question they disagree directly.</em></p>

<h2 id="how-to-read-this">How to read this</h2>

<p>Two axes are graded separately, per claim — not per section:</p>

<ul>
  <li><strong>Confidence</strong> — how good the sourcing and the reasoning are: <strong>High</strong> (corroborated by two or more
independent authoritative sources), <strong>Moderate</strong> (one credible source, or corroboration with gaps),
<strong>Low</strong> (single-source, self-claimed, or heavily inferential).</li>
  <li><strong>Probability</strong> — how likely the assessed outcome is, stated only where the claim is estimative:
<em>nearly certain</em>, <em>very likely</em>, <em>likely</em>, <em>roughly even chance</em>, <em>unlikely</em>, <em>very unlikely</em>.</li>
</ul>

<p>They are not the same thing. A claim can be <em>very likely</em> at <em>low confidence</em> (a single source) or
<em>roughly even</em> at <em>high confidence</em> (strong data, genuine uncertainty). Purely factual claims are
graded for confidence only and carry <em>n/a</em> for probability. <strong>Unverified</strong> means it could not be
substantiated, and is recorded so it is not repeated as fact.</p>

<h2 id="bottom-line">Bottom line</h2>

<p>Grandoreiro is a Delphi banking trojan of Brazilian origin, active since at least 2016 and
operated as a service with the code held closely. The January 2024 arrests of five administrators did not stop it: campaigns in 2024, 2025 and
2026 are documented by three independent vendors, and the codebase split rather than disappeared.</p>

<p>The interesting question is not whether it survived. It is <strong>who runs it now</strong>. IBM counts at
least three distributors; ESET, looking at the command-and-control backend, argues for one group
or a few groups cooperating closely; Kaspersky describes a service that fragmented after the
arrests. Read the operator model as <strong>contested</strong>, not as settled — that is where this product
departs from the vendor summaries.</p>

<h2 id="1-what-grandoreiro-is">1. What Grandoreiro is</h2>

<p>A Windows banking trojan written in Delphi, Brazilian in origin and part of the family Kaspersky
calls the <strong>Tetrade</strong>, alongside Guildma, Javali and Melcoz [3][6]. MITRE records it as software
S0531, first observed in 2016 and using a malware-as-a-service model [7]. INTERPOL dates the
campaign against Spanish-speaking victims to 2017 [1]; the two are not in conflict — 2016 is the
first sample, 2017 the start of sustained targeting.</p>

<p>Its purpose is account takeover, not mass encryption. It watches the foreground window for a
banking site, overlays it, captures keystrokes and one-time codes, and lets an operator drive the
victim’s machine to move money [2][3][4]. It also harvests addresses from the local Outlook
profile and uses the infected mailbox to send further phishing [3][4]. Older builds padded
themselves past 300 MB with bitmap resources to defeat scanning; that technique was dropped in
2023 [2][3].</p>

<p>The service model is unusual: Kaspersky found no underground advertisement for the Grandoreiro
package, and access to the source and builders appears limited to trusted partners [3]. That is
one of the reasons the operator question below is hard: a closed service and a small criminal
group look similar from the outside.</p>

<h2 id="2-assessment-ledger">2. Assessment ledger<!-- claims:begin --></h2>

<table>
  <thead>
    <tr>
      <th>Assessment</th>
      <th>Confidence</th>
      <th>Probability</th>
      <th>Basis</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>A Delphi banking trojan of Brazilian origin, part of the Tetrade family</td>
      <td>High</td>
      <td>n/a</td>
      <td>[3][6][7]</td>
    </tr>
    <tr>
      <td>Active since at least 2016, and tracked against Spanish-speaking victims since 2017</td>
      <td>High</td>
      <td>n/a</td>
      <td>[3][1]</td>
    </tr>
    <tr>
      <td>Operated as a service, with access to the source and builders limited to trusted partners</td>
      <td>Moderate</td>
      <td>n/a</td>
      <td>[3][4]</td>
    </tr>
    <tr>
      <td>IBM tracks at least three distributors: Hive0148, Hive0149 and a third in development</td>
      <td>Moderate</td>
      <td>n/a</td>
      <td>[5]</td>
    </tr>
    <tr>
      <td>After the 2024 arrests the codebase split: a lighter variant focused on Mexico runs alongside the main one</td>
      <td>Moderate</td>
      <td>n/a</td>
      <td>[3][6]</td>
    </tr>
    <tr>
      <td>Targets more than 1,500 banking applications; the country count differs by vendor (45 in Kaspersky, 60+ in IBM)</td>
      <td>High</td>
      <td>n/a</td>
      <td>[3][4]</td>
    </tr>
    <tr>
      <td>In January 2024, five programmers and operators were arrested in Brazil, in an operation coordinated by INTERPOL with Spain</td>
      <td>Moderate</td>
      <td>n/a</td>
      <td>[1][2]</td>
    </tr>
    <tr>
      <td>Operations continued after the arrests: campaigns in 2024, 2025 and 2026 are documented by three independent vendors</td>
      <td>High</td>
      <td>n/a</td>
      <td>[3][5][6]</td>
    </tr>
    <tr>
      <td>The 2026 campaign used a DLL sideloading chain through Duplicate Files Finder, concentrated in Mexico</td>
      <td>Moderate</td>
      <td>n/a</td>
      <td>[6]</td>
    </tr>
    <tr>
      <td>At least EUR 3.5 million stolen in Spain, with more than EUR 110 million attempted per CaixaBank</td>
      <td>Moderate</td>
      <td>n/a</td>
      <td>[1]</td>
    </tr>
    <tr>
      <td>Grandoreiro is run by several distributors sharing a codebase, not one organisation</td>
      <td>Moderate</td>
      <td>roughly even chance</td>
      <td>[5][3]</td>
    </tr>
    <tr>
      <td>Grandoreiro is run by one group or a few closely cooperating groups</td>
      <td>Moderate</td>
      <td>roughly even chance</td>
      <td>[2]</td>
    </tr>
    <tr>
      <td>Grandoreiro is a vendor label applied to unrelated malware</td>
      <td>Low</td>
      <td>very unlikely</td>
      <td>[3][6]</td>
    </tr>
  </tbody>
</table>

<!-- claims:end -->

<h2 id="3-the-operator-question--three-readings">3. The operator question — three readings</h2>

<p>Vendors agree on the malware and disagree on the organisation. Three readings survive long enough
to test:</p>

<table>
  <thead>
    <tr>
      <th>Reading</th>
      <th>What it predicts</th>
      <th>What the record shows</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td><strong>One group, or a few groups cooperating closely</strong></td>
      <td>A shared backend, load-balanced command and control, no independent forks</td>
      <td>ESET found the backend admits one operator at a time, and 94% of observed victims sat in a single DGA cluster — its explicit conclusion was one group or a few cooperating [2]</td>
    </tr>
    <tr>
      <td><strong>Several distributors on a shared codebase</strong></td>
      <td>Divergent loaders, C2 and phishing themes; forks after disruption</td>
      <td>IBM tracks at least three distributors (Hive0148, Hive0149, one in development) separated by loader, C2 and theme [5]; Kaspersky found two codebases running in parallel after the arrests [3]</td>
    </tr>
    <tr>
      <td><strong>A label over unrelated malware</strong></td>
      <td>No shared code between variants carrying the name</td>
      <td>Shared string decryption, DGA design and Portuguese C2 tokens appear across variants [3][4][6] — this reading fails</td>
    </tr>
  </tbody>
</table>

<p>Both of the first two are supported by direct observation, and they are not the same claim. ESET’s
argument is structural: if the same backend serialises operators and most victims land on one
cluster, several independent distributors are hard to sustain. IBM’s is behavioural: three
distinct infection chains, loaders and themes are hard to sustain as one team. The honest grade
is that they are roughly evenly matched on public evidence, with the balance turning on whether
you weight backend architecture or distribution tradecraft.</p>

<p>What would settle it is not another summary but an artefact: shared infrastructure or shared code
between two named distributor clusters. Neither vendor publishes that link.</p>

<h2 id="4-tradecraft">4. Tradecraft</h2>

<p><strong>Delivery.</strong> Phishing that impersonates the tax authority, the electricity utility, a court or a
telecom, in the target’s language, with a geofenced redirect so that only victims in the intended
country receive the payload [1][4][5]. The archive carries a loader, often with a decoy PDF and
XML inside [6]. Over the years the first stage has been an MSI, a VBS dropper [5] and, in the
2026 campaign, a DLL sideloading chain [6].</p>

<p><strong>The 2026 loader.</strong> Acronis documented a campaign in May 2026 that abuses the legitimate
<strong>Duplicate Files Finder</strong> application, renamed at random: the signed executable loads
<code class="language-plaintext highlighter-rouge">dupfdll.dll</code>, which loads a malicious <code class="language-plaintext highlighter-rouge">mingwm10.dll</code> from the same directory. The implant hides
the application window, checks system uptime, blacklists 49 analysis tools and looks for VMware
and VirtualBox artefacts before it contacts anything [6].</p>

<p><strong>Anti-analysis.</strong> Uptime thresholds, a long process blacklist, VM driver and BIOS checks,
desktop-shortcut profiling, country blacklists (Russia, Czechia, the Netherlands in one sample)
and a fake “error opening document” dialog in Spanish when a sandbox is suspected [4][6].</p>

<p><strong>Command and control.</strong> A domain generation algorithm rather than fixed infrastructure. IBM
counted 14 seeds in one sample, yielding 14 possible C2 domains per day, with seeds separated by
function — which is also how a service would keep its operators’ traffic apart [4]. Resolution
goes through Google’s DNS-over-HTTPS endpoint, and the port is derived from the resolved IP [4][6].
The C2 protocol speaks Portuguese: <code class="language-plaintext highlighter-rouge">CLIENT_SOLICITA_DDS_MODL</code>, <code class="language-plaintext highlighter-rouge">INFECTADO</code>, <code class="language-plaintext highlighter-rouge">AV_COMEU_MODULO</code> [4][6].</p>

<p><strong>Post-exploitation.</strong> Screen blocking and overlays to capture OTPs, remote mouse and keyboard
control, keylogging, webcam control, clipboard replacement for crypto wallets, file transfer, and
detection of banking security products such as Trusteer and Topaz OFD [2][3][4]. Persistence is a
registry Run key plus encrypted <code class="language-plaintext highlighter-rouge">.CFG</code> and <code class="language-plaintext highlighter-rouge">.XML</code> files [4].</p>

<h2 id="5-targeting">5. Targeting</h2>

<p>Brazil, Mexico and Spain are the long-standing core. ESET’s telemetry put Brazil at roughly 41% of
victims, Mexico at 30% and Spain at 28%, with a clear shift towards Mexico and Argentina in 2023
[2]. INTERPOL describes the operation as a threat across Spanish-speaking countries since 2017,
with money moved through mule networks and laundered back to Brazil [1]. Kaspersky measured the
target list growing from 900 banks in 40 countries in 2023 to 1,700 banks and 276 crypto wallets
in 45 countries in 2024 [3]; IBM, on its own sample set, reports more than 1,500 banking
applications in more than 60 countries [4]. The two counts are not reconcilable from public data,
and the difference is a reminder that these numbers describe a hardcoded list, not observed
victims.</p>

<p>IBM also documented expansion beyond the historical Brazil–Mexico–Spain core after the arrests
— South Africa, Argentina and the Indo-Pacific among the impersonated authorities [4]. Acronis’s 2026 telemetry is more
conservative: Mexico accounts for the largest share, Spain and several Latin American countries
follow, and Europe and North America show only limited presence [6].</p>

<h2 id="6-disruption-what-the-arrests-changed">6. Disruption: what the arrests changed</h2>

<p>In January 2024, Brazilian authorities arrested five programmers and operators, with searches
across five states, in an operation coordinated by INTERPOL and supported by Spain. Trend Micro,
Kaspersky, Group-IB and Scitum contributed analysis; ESET supplied technical analysis, statistics
and C2 indicators [1][2]. INTERPOL puts the Spanish losses at more than EUR 3.5 million, with
failed attempts that could have exceeded EUR 110 million according to CaixaBank [1].</p>

<p>What changed is visible; what it meant is not. Kaspersky observed a surge of Grandoreiro phishing
days after the arrests, and then two codebases in parallel — the updated one and a legacy build
aimed at about 30 Mexican banks, which it reads as a different developer working from older
source [3]. IBM saw Hive0148 campaigns in February–March 2025 against Mexico and Costa Rica [5].
Acronis saw a renewed campaign in May 2026 [6]. <strong>Degraded and fragmented, not dismantled.</strong></p>

<p>I found no public record of a conviction of the arrested administrators. That is an absence of
evidence, not evidence that none happened: Brazilian and Spanish proceedings are not fully
visible in English-language open sources, and this product does not claim to have searched court
records.</p>

<h2 id="7-what-this-means-for-defenders">7. What this means for defenders</h2>

<p>The target is a customer’s Windows endpoint, so the controls that matter are not the ones that
stop an intrusion into a corporate network:</p>

<ol>
  <li><strong>Break the loader chain.</strong> Block executables launched from archive downloads and from
user-writable paths, and alert on signed applications loading DLLs from their own directory —
the 2026 chain depends on exactly that.</li>
  <li><strong>Watch the two reconnaissance calls.</strong> <code class="language-plaintext highlighter-rouge">ip-api.com</code> and <code class="language-plaintext highlighter-rouge">dns.google/resolve</code> in quick
succession, followed by an outbound connection on an unusual high port, is a high-signal
sequence for this family [4][6].</li>
  <li><strong>Treat Outlook automation as suspicious.</strong> A workstation harvesting addresses and sending
mail at volume is not normal user behaviour [3][4].</li>
  <li><strong>Harden the transaction, not only the login.</strong> Overlay and OTP capture defeat password and
SMS factors; transaction signing, device binding and out-of-band confirmation of the payee are
the controls that survive an infected endpoint.</li>
  <li><strong>Block known DGA-derived domains, but do not rely on it.</strong> The DGA rotates daily and seeds
change [4].</li>
</ol>

<h2 id="8-detection-opportunities">8. Detection opportunities</h2>

<p>What to look for, not a validated ruleset:</p>

<ul>
  <li>A renamed copy of Duplicate Files Finder loading <code class="language-plaintext highlighter-rouge">mingwm10.dll</code> from the same directory [6].</li>
  <li>A GUI process that hides its window, then queries <code class="language-plaintext highlighter-rouge">ip-api.com</code> and the BIOS registry key in
quick succession [6].</li>
  <li>A DNS-over-HTTPS lookup to <code class="language-plaintext highlighter-rouge">dns.google/resolve?name=</code> immediately followed by TCP/6432 or
another non-standard port [4][6].</li>
  <li>The token <code class="language-plaintext highlighter-rouge">CLIENT_SOLICITA_DDS_MODL</code> in egress traffic or process memory [4][6].</li>
  <li>Registry Run keys pointing at an executable under <code class="language-plaintext highlighter-rouge">C:\ProgramData\</code>, with encrypted <code class="language-plaintext highlighter-rouge">.CFG</code> and
<code class="language-plaintext highlighter-rouge">.XML</code> files alongside [4].</li>
  <li>Outlook sending mail to many recipients from a single workstation, outside any marketing
process [3][4].</li>
</ul>

<h2 id="9-intelligence-gaps-and-what-i-could-not-verify">9. Intelligence gaps and what I could not verify</h2>

<ul>
  <li><strong>The operator model.</strong> The central question is contested and the two camps do not publish the
artefact that would resolve it.</li>
  <li><strong>The counts.</strong> 1,500 applications in 60+ countries (IBM) versus 1,700 in 45 (Kaspersky) cannot
both be right as stated; neither vendor publishes its sample set.</li>
  <li><strong>Distributor boundaries.</strong> Hive0148 and Hive0149 are IBM’s clusters, defined by loader, C2 and
theme. They are analytic constructs, not observed organisations.</li>
  <li><strong>The 2026 campaign’s owner.</strong> Acronis describes the campaign, not which distributor ran it.</li>
  <li><strong>Legal outcome.</strong> No public conviction found in English-language sources.</li>
  <li><strong>Method limit.</strong> This is an open-source profile. I did not analyse samples or use private
telemetry, and I did not exhaust Portuguese- or Spanish-language court reporting.</li>
</ul>

<h2 id="10-outlook--what-would-change-this-judgement">10. Outlook — what would change this judgement</h2>

<ul>
  <li>A <strong>conviction or indictment</strong> naming operators, which would turn the arrest claim from a
disruption event into an organisational finding.</li>
  <li><strong>Shared infrastructure or code</strong> linking two named distributor clusters — the single strongest
lever on the contested question.</li>
  <li>A <strong>vendor merging or splitting</strong> the distributor clusters, or ESET revising its single-group
assessment against new backend evidence.</li>
  <li>A <strong>new codebase</strong> that abandons the legacy one, or a third fork, which would date the
post-arrest split.</li>
  <li>A <strong>new country wave</strong> outside the current footprint, or a sustained pause longer than two
quarters.</li>
</ul>

<h2 id="sources">Sources</h2>

<ol>
  <li>INTERPOL — <strong>Disrupting a Grandoreiro malware operation</strong> (18 Mar 2024).
<a href="https://www.interpol.int/en/News-and-Events/News/2024/Disrupting-a-Grandoreiro-malware-operation">https://www.interpol.int/en/News-and-Events/News/2024/Disrupting-a-Grandoreiro-malware-operation</a></li>
  <li>ESET Research — <strong>ESET takes part in global operation to disrupt the Grandoreiro banking
trojan</strong> (30 Jan 2024).
<a href="https://www.welivesecurity.com/en/eset-research/eset-takes-part-global-operation-disrupt-grandoreiro-banking-trojan/">https://www.welivesecurity.com/en/eset-research/eset-takes-part-global-operation-disrupt-grandoreiro-banking-trojan/</a></li>
  <li>Kaspersky GReAT — <strong>Grandoreiro, the global trojan with grandiose goals</strong> (22 Oct 2024).
<a href="https://securelist.com/grandoreiro-banking-trojan/114257/">https://securelist.com/grandoreiro-banking-trojan/114257/</a></li>
  <li>IBM X-Force — <strong>Grandoreiro banking trojan unleashed: X-Force observing emerging global
campaigns</strong> (2024).
<a href="https://www.ibm.com/think/x-force/grandoreiro-banking-trojan-unleashed">https://www.ibm.com/think/x-force/grandoreiro-banking-trojan-unleashed</a></li>
  <li>IBM X-Force — <strong>Hive0148 observed targeting Mexico and Costa Rica</strong> (2025).
<a href="https://www.ibm.com/think/x-force/ibm-x-force-threat-analysis-hive0148-observed-targeting-mexico-costa-rica">https://www.ibm.com/think/x-force/ibm-x-force-threat-analysis-hive0148-observed-targeting-mexico-costa-rica</a></li>
  <li>Acronis TRU — <strong>Grandoreiro goes north: from Brazil to Mexico with a new DLL sideloading
campaign</strong> (19 Aug 2026).
<a href="https://www.acronis.com/en/tru/posts/grandoreiro-goes-north-from-brazil-to-mexico-with-a-new-dll-sideloading-campaign/">https://www.acronis.com/en/tru/posts/grandoreiro-goes-north-from-brazil-to-mexico-with-a-new-dll-sideloading-campaign/</a></li>
  <li>MITRE ATT&amp;CK — <strong>Software S0531: Grandoreiro</strong> (last modified 11 Apr 2024).
<a href="https://attack.mitre.org/software/S0531">https://attack.mitre.org/software/S0531</a></li>
</ol>

<hr />

<p><em>All analysis is open-source. Views are my own.</em></p>

<h2 id="revision-history">Revision history</h2>

<table>
  <thead>
    <tr>
      <th>Version</th>
      <th>Date</th>
      <th>Change</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>1.0</td>
      <td>2026-10-29</td>
      <td>First release.</td>
    </tr>
    <tr>
      <td>1.0.1</td>
      <td>2026-10-29</td>
      <td>Corrected a geographic error: Argentina is in Latin America, so the expansion is beyond the historical core, not outside the region. Also tightened the wording of the arrests. No assessed fact changed.</td>
    </tr>
  </tbody>
</table>]]></content><author><name>Martín Eliseo</name></author><category term="threat-intelligence" /><summary type="html"><![CDATA[A Brazilian banking trojan run as a service, measured against what the arrests changed — and the operator model the vendors still contest.]]></summary></entry><entry><title type="html">The broker is the actor you can see: SYLVANITE and the alias problem at the access layer</title><link href="https://martineliseo.es/sylvanite-access-layer/" rel="alternate" type="text/html" title="The broker is the actor you can see: SYLVANITE and the alias problem at the access layer" /><published>2026-10-28T00:00:00+00:00</published><updated>2026-10-28T00:00:00+00:00</updated><id>https://martineliseo.es/sylvanite-access-layer</id><content type="html" xml:base="https://martineliseo.es/sylvanite-access-layer/"><![CDATA[<table>
  <thead>
    <tr>
      <th> </th>
      <th> </th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td><strong>Product</strong></td>
      <td>Assessment (access layer, with a hunt plan)</td>
    </tr>
    <tr>
      <td><strong>Handling</strong></td>
      <td>TLP:CLEAR — open sources only</td>
    </tr>
    <tr>
      <td><strong>Version</strong></td>
      <td>1.1 (2026-10-28)</td>
    </tr>
    <tr>
      <td><strong>Review trigger</strong></td>
      <td>A vendor or government statement linking the access clusters; a new edge-device campaign; a public incident that shows the handoff</td>
    </tr>
    <tr>
      <td><strong>Next review</strong></td>
      <td>2027-04-30</td>
    </tr>
    <tr>
      <td><strong>Author</strong></td>
      <td>Martín Eliseo</td>
    </tr>
  </tbody>
</table>

<p><em>Every claim below carries a source and is graded on two independent axes. Where vendors disagree, I say so.</em></p>

<h2 id="bottom-line">Bottom line</h2>

<p>Three judgements, in the order a defender needs them:</p>

<ol>
  <li><strong>The access layer is tracked at three levels that are not interchangeable</strong> — a MITRE <em>campaign</em>
(C0029, Cutting Edge), a Dragos <em>group</em> (SYLVANITE) and Mandiant <em>clusters</em> (UNC5221, UNC5174,
UNC5291). Flattening them into an alias list deletes the broker, which is the only part a defender
can see before the OT intrusion starts.</li>
  <li><strong>The handoff is a direct observation, not an inference from shared tradecraft.</strong> Dragos states it
observed SYLVANITE transferring access to VOLTZITE, with the interval as short as days [1]. That is
a different class of evidence from the vendor language around identity, and it is enough to support
a hunt.</li>
  <li><strong>The broker is still not one entity.</strong> Dragos assesses the names it lists as overlapping SYLVANITE
as <em>multiple entities under the same overarching direction</em> [1]. The honest unit of analysis is an
<strong>access ecosystem</strong>, not a broker — and the hunt below is written for the ecosystem.</li>
</ol>

<h2 id="the-decision-this-answers">The decision this answers</h2>

<p><a href="/requirements/">PIR-1</a> asks which actors pose a credible disruptive intent against OT; PIR-3 asks
where vendor designations diverge and how much confidence the evidence supports. This assessment sits
where they meet, because the access layer is where a divergence has an operational cost: if a
SYLVANITE compromise is the leading edge of a VOLTZITE intrusion, then an edge-device incident is an
OT-scope event, and the response clock starts at the appliance, not at the engineering workstation.</p>

<h2 id="three-levels-and-why-they-are-not-the-same-object">Three levels, and why they are not the same object</h2>

<table>
  <thead>
    <tr>
      <th>Level</th>
      <th>What it is</th>
      <th>Who records it</th>
      <th>What it is <em>not</em></th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td><strong>Campaign C0029</strong></td>
      <td>Ivanti Connect Secure zero-day exploitation, Dec 2023 – Feb 2024, attributed variously to UNC5221/UTA0178 and UNC5325 [3]</td>
      <td>MITRE ATT&amp;CK</td>
      <td>Not a group, and not a name for one</td>
    </tr>
    <tr>
      <td><strong>SYLVANITE</strong></td>
      <td>A Dragos-tracked initial-access group, active since 2023, that weaponises edge-device proofs of concept and hands footholds to Stage 2 actors [1]</td>
      <td>Dragos</td>
      <td>Not an assertion that its member clusters are one organisation</td>
    </tr>
    <tr>
      <td><strong>UNC5221 / UNC5174 / UNC5291</strong></td>
      <td>Mandiant cluster designations, some with a persona attached (UNC5174 = “Uteus”) [2]</td>
      <td>Mandiant / Google</td>
      <td>Not campaign names, and not evidence of a single operator</td>
    </tr>
  </tbody>
</table>

<p>A campaign is an <em>event with a window</em>. A group is a <em>tracking construct</em>. A cluster is a <em>set of
activity with a shared toolset</em>. Citing one for another is how this layer acquired a false clarity:
the published Volt Typhoon profile did exactly that, and it was corrected on 2026-10-28 (profile v2.2).</p>

<h2 id="what-was-actually-observed">What was actually observed</h2>

<p><strong>By Dragos.</strong> SYLVANITE monitors vulnerability research and weaponises newly disclosed exploits,
in several cases compromising targets before the vendor had issued a patch. Its toolset includes
memory-resident web shells (Godzilla, LIGHTWIRE, WIREFIRE), Cobalt Strike, Sliver and SUPERSHELL, and
FRP for tunnelling; it harvests credentials from backend systems, including LDAP data and Office 365
tokens. Dragos states that it directly observed the group transferring access to VOLTZITE, and that
the handoff can occur within days of initial access [1].</p>

<p><strong>By Mandiant.</strong> UNC5174, using the persona “Uteus”, exploited F5 BIG-IP CVE-2023-46747 and
ConnectWise ScreenConnect CVE-2024-1709, and was observed <em>attempting to sell</em> access to U.S. defense
contractor and UK government victims. Mandiant assesses with moderate confidence that UNC5174 is a
PRC access broker with a claimed MSS affiliation [2]. One behaviour is worth keeping: after taking
access, UNC5174 ran the vendor’s own mitigation script and removed an unrelated actor’s account —
closing the door behind itself. Access was being treated as an asset, which is what a broker does.</p>

<p><strong>By MITRE.</strong> C0029 records the campaign, its techniques (T1190, T1505.003, T1078.002, T1003.003,
T1572 among them) and its software (WIREFIRE, LIGHTWIRE, GLASSTOKEN) [3]. It does not record a group
called SYLVANITE — ATT&amp;CK v20 carries 193 enterprise groups and none is SYLVANITE [4].</p>

<h2 id="the-name-graph-read-flatly">The name graph, read flatly</h2>

<div class="language-plaintext highlighter-rouge"><div class="highlight"><pre class="highlight"><code>MITRE (who is named)              Dragos (what is tracked)         Mandiant (what was seen)
─────────────────────             ────────────────────────         ────────────────────────
Volt Typhoon ──alias── UNC3236
        │                              ⋮ overlap
        │                         SYLVANITE ──overlap── UNC5221 / UTA0178 ──► campaign C0029
        │                              │                                  UNC5325
        │                              ├──overlap── UNC5174 ("Uteus")
        │                              ├──overlap── UNC5291
        │                              └──overlap── HOUKEN / Red Dev 61 / CL-STA-0048
        └───────────── handoff observed ──────────────┘
</code></pre></div></div>

<p>The trap is one edge. Dragos lists <strong>UNC3236</strong> among the names overlapping SYLVANITE [1]. ATT&amp;CK
lists UNC3236 as an alias of Volt Typhoon [4]. Read with the flat alias convention most platforms
use — two names in one field are two names for one thing — UNC3236 is Volt Typhoon, UNC3236 overlaps
SYLVANITE, therefore SYLVANITE <em>is</em> Volt Typhoon, and the broker disappears from the graph.</p>

<p>No vendor asserts that. The record says “overlaps”, which is a statement about shared activity, not
about shared organisation. This is the same error the <a href="/ach-voltzite-volt-typhoon/">ACH</a> was written
to avoid, one layer down.</p>

<h2 id="assessment-ledger">Assessment ledger<!-- claims:begin --></h2>

<table>
  <thead>
    <tr>
      <th>Assessment</th>
      <th>Confidence</th>
      <th>Probability</th>
      <th>Basis</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>SYLVANITE is a Dragos-tracked initial-access group active since 2023 that weaponises edge-device proofs of concept</td>
      <td>Moderate</td>
      <td>n/a</td>
      <td>[1]</td>
    </tr>
    <tr>
      <td>Dragos states it directly observed SYLVANITE handing established footholds to VOLTZITE, with the interval as short as days</td>
      <td>Moderate</td>
      <td>n/a</td>
      <td>[1]</td>
    </tr>
    <tr>
      <td>The names overlapping SYLVANITE are best read as multiple entities under one direction, not one actor</td>
      <td>Moderate</td>
      <td>likely</td>
      <td>[1]</td>
    </tr>
    <tr>
      <td>UNC5174 (Uteus) is assessed as a PRC access broker that attempted to sell access to edge-device victims</td>
      <td>Moderate</td>
      <td>likely</td>
      <td>[2]</td>
    </tr>
    <tr>
      <td>ATT&amp;CK records C0029 as a campaign covering Ivanti exploitation, not a group named SYLVANITE</td>
      <td>Moderate</td>
      <td>n/a</td>
      <td>[3][4]</td>
    </tr>
    <tr>
      <td>UNC3236 is both an ATT&amp;CK alias of Volt Typhoon and a Dragos overlap of SYLVANITE</td>
      <td>High</td>
      <td>n/a</td>
      <td>[1][4]</td>
    </tr>
    <tr>
      <td>SYLVANITE and UNC5174 are the same actor</td>
      <td>Low</td>
      <td>unlikely</td>
      <td>[1][2]</td>
    </tr>
  </tbody>
</table>

<!-- claims:end -->

<p>The last two rows are the ones to hold carefully. Row 7 is <em>not</em> “unlikely” because the clusters look
different; it is “unlikely” because the only thing joining them is commodity tooling — SUPERSHELL is
used by the administrators of the framework as much as by any one customer — and because the tracking
vendor that knows the cluster best says the names are several entities. Row 8 is a judgement about
consequence, not about identity: it survives even if every cluster in the list turns out to be
separate.</p>

<h2 id="hunt-plan">Hunt plan</h2>

<p>The plan is written to be refuted. It is stored at
<a href="/hunts/sylvanite-handoff.json"><code class="language-plaintext highlighter-rouge">hunts/sylvanite-handoff.json</code></a> and validated with <code class="language-plaintext highlighter-rouge">huntplan</code> 0.2.0
(<code class="language-plaintext highlighter-rouge">ready: true</code>, no findings).</p>

<p><strong>Hypothesis.</strong> If SYLVANITE handed access to VOLTZITE, then edge-appliance audit logs will show a
newly created local administrator account followed within 14 days by domain-account logons and
credential dumping on the same estate.</p>

<p><strong>Anchors</strong> — the plan is only threat-informed if it names techniques: T1078 (valid accounts),
T1021 (remote services), T1003 (OS credential dumping), T1059 (command and scripting interpreter).</p>

<p><strong>What a hit looks like</strong>, using the detection strategies the suite maps for those techniques:</p>

<ul>
  <li>a local account created on a VPN/SD-WAN appliance or firewall, and the same identity authenticating
to the domain (T1078; logon-session and account-metadata components);</li>
  <li>remote service use from that identity — RDP, SMB, SSH, WinRM (T1021; logon-session creation plus
process creation on the destination);</li>
  <li>credential dumping on a host reached that way (T1003; process access, file access, LSASS/NTDS
analytics);</li>
  <li>shell or scripting execution that does not match the administrative baseline on those hosts (T1059;
process creation, command execution).</li>
</ul>

<p><strong>Closure.</strong> Confirmed if at least two incidents show the three stages in order within 14 days.
Refuted if, across three public incident reports or 90 days of telemetry, broker-stage artefacts are
absent in the majority, or the OT-facing stage systematically precedes them. A negative result closes
the hunt for that window; it does not establish that the actor is absent.</p>

<p><strong>One honest limit.</strong> <code class="language-plaintext highlighter-rouge">huntplan</code> 0.2.0 ships a curated detection-strategy subset of 12 techniques and
does not include T1190, so the plan anchors the post-access stages it can map. The edge stage is
covered by the appliance’s own audit log, which is where the hypothesis starts anyway.</p>

<h2 id="what-would-change-this-judgement">What would change this judgement</h2>

<ul>
  <li>A vendor or government statement that the clusters are one organisation, or that the handoff is a
sale rather than a tasking relationship.</li>
  <li>Published infrastructure or code shared <em>between</em> the access clusters, as opposed to tooling
shared across the ecosystem.</li>
  <li>Evidence that SYLVANITE serves more than one consumer cluster, which would confirm the broker model
and weaken any single-actor reading.</li>
  <li>A disruptive effect attributed to this chain: that would move the whole assessment from access to
consequence.</li>
</ul>

<h2 id="intelligence-gaps">Intelligence gaps</h2>

<ul>
  <li>The handoff is a single-vendor observation. No government advisory names SYLVANITE.</li>
  <li>“Access ecosystem” is a construct I am imposing to avoid a worse one; no source uses the term.</li>
  <li>Mandiant’s UNC5174 assessment is from March 2024; the persona may have moved, been absorbed, or
changed customers since.</li>
  <li>No telemetry, no infrastructure analysis, no malware samples: this is an open-source assessment and
is bounded by that.</li>
  <li>Whether the same handoff pattern holds outside U.S. electric and water utilities is unknown.</li>
</ul>

<h2 id="sources">Sources</h2>

<ol>
  <li>Dragos — <strong>SYLVANITE</strong> threat group page (accessed 28 Oct 2026).
<a href="https://www.dragos.com/threat/sylvanite/">https://www.dragos.com/threat/sylvanite/</a></li>
  <li>Mandiant — <strong>Bringing Access Back — Initial Access Brokers Exploit F5 BIG-IP (CVE-2023-46747) and
ScreenConnect</strong> (21 Mar 2024).
<a href="https://cloud.google.com/blog/topics/threat-intelligence/initial-access-brokers-exploit-f5-screenconnect">https://cloud.google.com/blog/topics/threat-intelligence/initial-access-brokers-exploit-f5-screenconnect</a></li>
  <li>MITRE ATT&amp;CK — <strong>Campaign C0029: Cutting Edge</strong> (last modified 31 Jul 2026).
<a href="https://attack.mitre.org/campaigns/C0029">https://attack.mitre.org/campaigns/C0029</a></li>
  <li>MITRE ATT&amp;CK — <strong>Group G1017: Volt Typhoon</strong> (v2.0, last modified 31 Jul 2026).
<a href="https://attack.mitre.org/groups/G1017">https://attack.mitre.org/groups/G1017</a></li>
  <li>CISA, NSA, FBI and international partners — <strong>Joint Cybersecurity Advisory AA24-038A</strong> (7 Feb 2024).
<a href="https://www.cisa.gov/news-events/cybersecurity-advisories/aa24-038a">https://www.cisa.gov/news-events/cybersecurity-advisories/aa24-038a</a></li>
</ol>

<hr />

<p><em>All analysis is open-source. Views are my own.</em></p>

<h2 id="revision-history">Revision history</h2>

<table>
  <thead>
    <tr>
      <th>Version</th>
      <th>Date</th>
      <th>Change</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>1.0</td>
      <td>2026-10-28</td>
      <td>First release.</td>
    </tr>
    <tr>
      <td>1.1</td>
      <td>2026-10-28</td>
      <td>The operational recommendation left the ledger: a recommendation is not a claim, and the ledger now holds only what evidence can grade. It remains in the bottom line and in the defender section. Grades are computed from the claim registry.</td>
    </tr>
  </tbody>
</table>]]></content><author><name>Martín Eliseo</name></author><category term="threat-intelligence" /><summary type="html"><![CDATA[The access layer as an actor: SYLVANITE, the alias problem, and what a broker looks like when the buyer is the story.]]></summary></entry><entry><title type="html">Is VOLTZITE the same actor as Volt Typhoon? Two questions, not three hypotheses</title><link href="https://martineliseo.es/ach-voltzite-volt-typhoon/" rel="alternate" type="text/html" title="Is VOLTZITE the same actor as Volt Typhoon? Two questions, not three hypotheses" /><published>2026-10-09T00:00:00+00:00</published><updated>2026-10-09T00:00:00+00:00</updated><id>https://martineliseo.es/ach-voltzite-volt-typhoon</id><content type="html" xml:base="https://martineliseo.es/ach-voltzite-volt-typhoon/"><![CDATA[<table>
  <thead>
    <tr>
      <th> </th>
      <th> </th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td><strong>Product</strong></td>
      <td>Assessment (structured technique: ACH, two questions)</td>
    </tr>
    <tr>
      <td><strong>Handling</strong></td>
      <td>TLP:CLEAR — open sources only</td>
    </tr>
    <tr>
      <td><strong>Version</strong></td>
      <td>2.1 (2026-10-29)</td>
    </tr>
    <tr>
      <td><strong>Review trigger</strong></td>
      <td>A vendor or government statement that explicitly asserts or denies identity; published infrastructure or code reuse; detail on SYLVANITE tasking</td>
    </tr>
    <tr>
      <td><strong>Next review</strong></td>
      <td>2027-04-30</td>
    </tr>
    <tr>
      <td><strong>Author</strong></td>
      <td>Martín Eliseo</td>
    </tr>
  </tbody>
</table>

<p><em>Every claim below carries a source and is graded on two independent axes. The grades are computed
from a registry (<code class="language-plaintext highlighter-rouge">_data/claims/</code>), not typed into the table.</em></p>

<h2 id="bottom-line">Bottom line</h2>

<ol>
  <li><strong>The identity question is undecided, and saying so is the finding.</strong> “Same unit” and “distinct
but related” both sit at <strong>roughly even</strong>, at <strong>low confidence</strong>. What the open evidence supports
is that <em>some</em> relationship exists rather than none — “unrelated” is <strong>very unlikely</strong>.</li>
  <li><strong>The mechanism is better evidenced than the identity.</strong> A separate cluster establishing access
and handing it to VOLTZITE is <strong>likely</strong>, at <strong>moderate confidence</strong>: that is a direct observation,
not an inference from shared tradecraft.</li>
  <li><strong>This version changes the identity judgement.</strong> v1.3 said “unlikely (20–45%)”. That conclusion
rested on counting one observation three times; counted once, the estimate returns to the band it
started from.</li>
</ol>

<h2 id="what-changed-from-v13-and-why">What changed from v1.3, and why</h2>

<table>
  <thead>
    <tr>
      <th>v1.3</th>
      <th>v2.0</th>
      <th>Why</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>One matrix: H1 same, H2 related, H3 brokered</td>
      <td>Two matrices: <strong>identity</strong> and <strong>mechanism</strong></td>
      <td>H3 was compatible with H2, so the matrix could never refute it. A hypothesis that cannot lose is not being tested.</td>
    </tr>
    <tr>
      <td>E1, E4 and E9 counted as three inconsistencies against H1</td>
      <td>One item (A-i), with its independence stated</td>
      <td>All three say the same thing: no vendor asserts identity. Three rows for one fact inflate the case.</td>
    </tr>
    <tr>
      <td>H2 defined broadly enough to absorb anything</td>
      <td>A2 has a refutation condition</td>
      <td>A hypothesis that explains every observation wins by construction.</td>
    </tr>
    <tr>
      <td>H1 <code class="language-plaintext highlighter-rouge">unlikely</code> (20–45%)</td>
      <td>A1 <code class="language-plaintext highlighter-rouge">roughly even</code> (45–55%)</td>
      <td>The triple count was what pushed it below the band. Grouped, it returns to the prior.</td>
    </tr>
    <tr>
      <td>H2 graded <code class="language-plaintext highlighter-rouge">moderate confidence</code></td>
      <td>A2 graded <code class="language-plaintext highlighter-rouge">low</code></td>
      <td>Nothing in the public evidence is diagnostic for identity: no infrastructure, no code, no operator artefact. Under the site’s ceiling, no diagnostic item means no more than low.</td>
    </tr>
  </tbody>
</table>

<h2 id="question-a--are-they-one-organisational-unit">Question A — are they one organisational unit?</h2>

<ul>
  <li><strong>A1 — Same unit.</strong> VOLTZITE and Volt Typhoon are one entity under two vendor names.</li>
  <li><strong>A2 — Distinct, related.</strong> Separate teams or units that share tooling, infrastructure, targeting or
tasking. <em>Refuted if</em> a vendor or government explicitly asserts identity, or published
infrastructure or code establishes a single operator.</li>
  <li><strong>A3 — Distinct, unrelated.</strong> The overlap is what any ICS-focused intrusion set looks like from
outside. <em>Refuted by</em> the same evidence that would confirm A1, or by a shared artefact.</li>
</ul>

<h3 id="evidence">Evidence</h3>

<table>
  <thead>
    <tr>
      <th>#</th>
      <th>Evidence</th>
      <th>Bears on</th>
      <th>Independence</th>
      <th>Diagnostic weight</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>A-i</td>
      <td>No vendor asserts identity: Dragos says “technical overlaps”, MITRE records Associated Groups, CISA lists an alias [3][7][12]</td>
      <td>Naming convention</td>
      <td>One observation across three publishers; two of them draw on vendor reporting</td>
      <td><strong>Weak</strong>: evidence about how names are assigned</td>
    </tr>
    <tr>
      <td>A-ii</td>
      <td>CISA lists Voltzite among Volt Typhoon’s aliases [1]</td>
      <td>Equivalence at campaign level</td>
      <td>Government, but an alias list is not an identity finding</td>
      <td><strong>Moderate</strong></td>
    </tr>
    <tr>
      <td>A-iii</td>
      <td>Same sectors, same living-off-the-land tradecraft, same SOHO proxying [1][10]</td>
      <td>Targeting and capability</td>
      <td>Several, independent</td>
      <td><strong>None</strong>: consistent with A1, A2 and A3</td>
    </tr>
    <tr>
      <td>A-iv</td>
      <td>No public shared infrastructure, no shared code, no operator artefact</td>
      <td>Identity</td>
      <td>—</td>
      <td><strong>Absence of evidence</strong>, not evidence of difference</td>
    </tr>
  </tbody>
</table>

<h3 id="matrix">Matrix</h3>

<p><code class="language-plaintext highlighter-rouge">C</code> consistent · <code class="language-plaintext highlighter-rouge">I</code> inconsistent · <code class="language-plaintext highlighter-rouge">0</code> neutral</p>

<table>
  <thead>
    <tr>
      <th>Evidence</th>
      <th>A1 same</th>
      <th>A2 related</th>
      <th>A3 unrelated</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>A-i naming language</td>
      <td>0</td>
      <td>C</td>
      <td>C</td>
    </tr>
    <tr>
      <td>A-ii CISA alias list</td>
      <td>C</td>
      <td>0</td>
      <td>0</td>
    </tr>
    <tr>
      <td>A-iii shared tradecraft</td>
      <td>C</td>
      <td>C</td>
      <td>C</td>
    </tr>
    <tr>
      <td>A-iv no linkage published</td>
      <td>0</td>
      <td>C</td>
      <td>C</td>
    </tr>
  </tbody>
</table>

<h3 id="judgement-a">Judgement A</h3>

<p>Only one item discriminates, and it points both ways: the government lists the names together, the
vendors decline to. That is not enough to separate A1 from A2, and it is not enough to exclude A3 on
open sources either.</p>

<h3 id="assessment-ledger">Assessment ledger<!-- claims:begin --></h3>

<p><strong>Question A — identity</strong></p>

<table>
  <thead>
    <tr>
      <th>Assessment</th>
      <th>Confidence</th>
      <th>Probability</th>
      <th>Basis</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>Dragos’s VOLTZITE is the same actor as Volt Typhoon</td>
      <td>Low</td>
      <td>roughly even chance</td>
      <td>[1][3][7][12]</td>
    </tr>
    <tr>
      <td>The clusters are distinct but related</td>
      <td>Low</td>
      <td>roughly even chance</td>
      <td>[1][3][7][12]</td>
    </tr>
    <tr>
      <td>The clusters are unrelated</td>
      <td>Low</td>
      <td>very unlikely</td>
      <td>[1][10]</td>
    </tr>
  </tbody>
</table>

<p><strong>Question B — mechanism</strong></p>

<table>
  <thead>
    <tr>
      <th>Assessment</th>
      <th>Confidence</th>
      <th>Probability</th>
      <th>Basis</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>SYLVANITE establishes footholds and hands them to VOLTZITE</td>
      <td>Moderate</td>
      <td>likely</td>
      <td>[12]</td>
    </tr>
    <tr>
      <td>VOLTZITE does its own initial access; the overlap is capability, not supply</td>
      <td>Low</td>
      <td>roughly even chance</td>
      <td>[9][10]</td>
    </tr>
    <tr>
      <td>Shared tradecraft without a standing handoff</td>
      <td>Low</td>
      <td>roughly even chance</td>
      <td>[12]</td>
    </tr>
  </tbody>
</table>

<!-- claims:end -->

<p>Both leading hypotheses are <code class="language-plaintext highlighter-rouge">low</code> because <strong>nothing diagnostic exists in public</strong>: the ceiling is set
by the evidence, and no amount of vendor language raises it. v1.3 graded “distinct but related” at
moderate confidence; that grade was not supported, and the registry refused it.</p>

<h2 id="question-b--is-the-overlap-produced-by-brokered-access">Question B — is the overlap produced by brokered access?</h2>

<ul>
  <li><strong>B1 — Brokered.</strong> A separate cluster establishes footholds and hands them to VOLTZITE.</li>
  <li><strong>B2 — Direct.</strong> VOLTZITE does its own initial access; the overlap is capability, not supply.</li>
  <li><strong>B3 — Shared tradecraft, no broker.</strong> Clusters exchange tooling or buy the same commodity access
without a standing handoff.</li>
</ul>

<h3 id="evidence-1">Evidence</h3>

<table>
  <thead>
    <tr>
      <th>#</th>
      <th>Evidence</th>
      <th>B1</th>
      <th>B2</th>
      <th>B3</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>B-i</td>
      <td>Dragos observed SYLVANITE weaponising edge-device vulnerabilities and handing established footholds to VOLTZITE, during incident response at U.S. electric and water utilities [12]</td>
      <td>C</td>
      <td>I</td>
      <td>0</td>
    </tr>
    <tr>
      <td>B-ii</td>
      <td>SYLVANITE is tracked as a separate group with its own overlaps — UNC5221, UNC5174, UNC5291, UNC3236 and others [12]</td>
      <td>C</td>
      <td>0</td>
      <td>C</td>
    </tr>
    <tr>
      <td>B-iii</td>
      <td>VOLTZITE has also exploited edge devices itself, e.g. Versa Director CVE-2024-39717 [9]</td>
      <td>0</td>
      <td>C</td>
      <td>C</td>
    </tr>
    <tr>
      <td>B-iv</td>
      <td>No public evidence of a sale, a price or a market transaction [12]</td>
      <td>0</td>
      <td>0</td>
      <td>0</td>
    </tr>
  </tbody>
</table>

<h3 id="judgement-b">Judgement B</h3>

<p><strong>B1 is likely, at moderate confidence.</strong> It is the only judgement here carried by direct observation
rather than by naming: Dragos states it watched the handoff. B2 is not excluded for other intrusions —
B-iii shows the actor still does its own initial access where it can — and B3 cannot be separated from
B1 on public evidence, because a one-directional handoff and a shared-access arrangement look the same
from outside.</p>

<p>The distinction matters for defenders: under B1, edge-device exploitation by <em>any</em> cluster in the
ecosystem is a precursor to this actor, not a separate problem to triage away.</p>

<h2 id="sensitivity--how-much-of-this-is-the-prior">Sensitivity — how much of this is the prior?</h2>

<p>The identity judgement is not robust, and it should say so.</p>

<p><strong>The prior.</strong> Before the evidence, the odds that two vendor cryptonyms for the same activity are one
organisational unit are close to even: renames are common, splits are common, and nothing here
establishes which this is.</p>

<p><strong>The discriminant.</strong> A-i is the only item that moves A1, and it is not about the actor: it is about
what vendors are willing to say. Counted once, it is weak. If the prior were 30% rather than even, no
combination of the current evidence would lift A1 out of “unlikely”. An explicit assertion of identity,
or published infrastructure or code establishing a shared operator, would move it across 50% on its
own.</p>

<p>Question B is different. B-i is a direct observation, so its judgement depends on whether Dragos is
right, not on a prior — which is why it carries moderate confidence with a single source and A1 does
not.</p>

<h2 id="what-would-break-the-tie">What would break the tie</h2>

<ul>
  <li>A vendor or government statement that <strong>explicitly asserts or denies identity</strong> — the alias list is
the weak form of this evidence, and it is already counted.</li>
  <li>Published <strong>infrastructure or code reuse</strong> establishing a shared operator rather than shared tooling.</li>
  <li>Detail on <strong>SYLVANITE tasking</strong>: whether it serves one consumer cluster or several.</li>
  <li>A joint advisory naming VOLTZITE and Volt Typhoon together as one actor or as distinct.</li>
</ul>

<h2 id="intelligence-gaps">Intelligence gaps</h2>

<ul>
  <li>Vendor taxonomy is internal and seldom explained; “overlap” is not a standardised term.</li>
  <li>The Dragos 2026 report is load-bearing for A-i and B-i and represents a single vendor.</li>
  <li>No public infrastructure or code links the two designations; this is the gap that keeps both leading
hypotheses at low confidence.</li>
  <li>No telemetry, no samples, no shared-infrastructure analysis: this is an open-source assessment and is
bounded by that.</li>
</ul>

<h2 id="sources">Sources</h2>

<p>The numbering follows the <a href="/volt-typhoon-actor-profile/">Volt Typhoon profile</a>, which holds the full
list; only the entries cited here are repeated. Entries 2, 4, 5, 6, 8 and 11 are in the profile and
are not cited in this product.</p>

<ol>
  <li>CISA, NSA, FBI and international partners — <strong>Joint Cybersecurity Advisory AA24-038A</strong> (7 Feb 2024).
<a href="https://www.cisa.gov/news-events/cybersecurity-advisories/aa24-038a">https://www.cisa.gov/news-events/cybersecurity-advisories/aa24-038a</a></li>
  <li>MITRE ATT&amp;CK — <strong>Group G1017: Volt Typhoon</strong> (v2.0, last modified 31 Jul 2026).
<a href="https://attack.mitre.org/groups/G1017">https://attack.mitre.org/groups/G1017</a></li>
  <li>Dragos — <strong>OT/ICS Cybersecurity Report 2026: Year in Review</strong> (17 Feb 2026).
<a href="https://www.dragos.com/resources/press-release/dragos-2026-year-in-review-new-ot-threats-ransomware">https://www.dragos.com/resources/press-release/dragos-2026-year-in-review-new-ot-threats-ransomware</a></li>
  <li>MITRE ATT&amp;CK — <strong>Campaign C0039: Versa Director Zero Day Exploitation</strong> (June–August 2024).
<a href="https://attack.mitre.org/campaigns/C0039">https://attack.mitre.org/campaigns/C0039</a></li>
  <li>Palo Alto Networks Unit 42 — <strong>Threat Brief: Attacks on Critical Infrastructure Attributed to
Insidious Taurus (Volt Typhoon)</strong> (Feb 2024).
<a href="https://unit42.paloaltonetworks.com/volt-typhoon-threat-brief/">https://unit42.paloaltonetworks.com/volt-typhoon-threat-brief/</a></li>
  <li>Dragos — <strong>SYLVANITE</strong> threat group page (accessed 28 Oct 2026).
<a href="https://www.dragos.com/threat/sylvanite/">https://www.dragos.com/threat/sylvanite/</a></li>
</ol>

<hr />

<p><em>All analysis is open-source. Views are my own.</em></p>

<h2 id="revision-history">Revision history</h2>

<table>
  <thead>
    <tr>
      <th>Version</th>
      <th>Date</th>
      <th>Change</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>1.0</td>
      <td>2026-10-09</td>
      <td>First release.</td>
    </tr>
    <tr>
      <td>1.1</td>
      <td>2026-10-28</td>
      <td>Handling corrected from the retired TLP 1.0 label to <code class="language-plaintext highlighter-rouge">TLP:CLEAR</code>, now checked in CI. The Volt Typhoon profile’s identity row was aligned to this product’s judgement. The matrix, the hypotheses and the judgement are unchanged.</td>
    </tr>
    <tr>
      <td>1.2</td>
      <td>2026-10-28</td>
      <td>Citation corrected: the SYLVANITE handoff (E7) is a Dragos observation, not an ATT&amp;CK record. The matrix, the hypotheses and the judgement are unchanged.</td>
    </tr>
    <tr>
      <td>1.3</td>
      <td>2026-10-28</td>
      <td>Added a sensitivity paragraph: H1 rests on a prior near even and one weak observation repeated three times; one documentary event would move it across 50%. The matrix and the judgement are unchanged.</td>
    </tr>
    <tr>
      <td>2.0</td>
      <td>2026-10-28</td>
      <td>Split identity from mechanism; grouped the naming observation; gave A2 a refutation condition; moved A1 from <code class="language-plaintext highlighter-rouge">unlikely</code> to <code class="language-plaintext highlighter-rouge">roughly even</code> and A2 from <code class="language-plaintext highlighter-rouge">moderate</code> to <code class="language-plaintext highlighter-rouge">low</code>; registered the judgements so the ceiling is computed. The identity judgement changed.</td>
    </tr>
    <tr>
      <td>2.1</td>
      <td>2026-10-29</td>
      <td>Question B is now its own ACH block in the registry and appears in the ledger, with B1’s probability aligned to the prose. The rendering was fixed so a product with two questions does not drop its second table. No assessed fact changed.</td>
    </tr>
  </tbody>
</table>]]></content><author><name>Martín Eliseo</name></author><category term="threat-intelligence" /><summary type="html"><![CDATA[Two separate questions — is VOLTZITE Volt Typhoon, and is the overlap brokered access? — tested with ACH instead of one unfalsifiable hypothesis.]]></summary></entry><entry><title type="html">From source to a defensible judgement: the method behind the tooling</title><link href="https://martineliseo.es/from-source-to-a-defensible-judgement/" rel="alternate" type="text/html" title="From source to a defensible judgement: the method behind the tooling" /><published>2026-10-09T00:00:00+00:00</published><updated>2026-10-09T00:00:00+00:00</updated><id>https://martineliseo.es/from-source-to-a-defensible-judgement</id><content type="html" xml:base="https://martineliseo.es/from-source-to-a-defensible-judgement/"><![CDATA[<table>
  <thead>
    <tr>
      <th> </th>
      <th> </th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td><strong>Product</strong></td>
      <td>Method note</td>
    </tr>
    <tr>
      <td><strong>Handling</strong></td>
      <td>TLP:CLEAR — open sources only</td>
    </tr>
    <tr>
      <td><strong>Version</strong></td>
      <td>1.1 (2026-10-28)</td>
    </tr>
    <tr>
      <td><strong>Review trigger</strong></td>
      <td>A change to the method, or a case that breaks it</td>
    </tr>
    <tr>
      <td><strong>Next review</strong></td>
      <td>2027-04-30</td>
    </tr>
    <tr>
      <td><strong>Author</strong></td>
      <td>Martín Eliseo</td>
    </tr>
  </tbody>
</table>

<h2 id="bottom-line">Bottom line</h2>

<p>A judgement is defensible when a reader can see <strong>why</strong> it was reached and <strong>what would
overturn it</strong>. This note describes the method I use to get there — and the tooling that
encodes it so the hard part cannot be skipped.</p>

<h2 id="the-problem-the-names-drift">The problem: the names drift</h2>

<p>Ask five vendors who an actor is and you may get five names. <em>Volt Typhoon</em>, <em>VOLTZITE</em>,
<em>Insidious Taurus</em>, <em>BRONZE SILHOUETTE</em>, <em>UNC3236</em>. Platforms then store those names as a
<strong>flat list of aliases</strong>, which quietly turns a hypothesis into a fact: if two names sit
together, they “are” the same actor. The reasoning is lost, the confidence is unstated,
and a decision — an incident response, a procurement choice, a threat hunt — is made on
ground that no longer shows why it holds.</p>

<p>I worked through one instance of this by hand in
<a href="/ach-voltzite-volt-typhoon/">Is VOLTZITE the same actor as Volt Typhoon?</a>. This note is
the step from doing it once to doing it repeatably.</p>

<h2 id="the-method">The method</h2>

<p>Four rules do the work.</p>

<p><strong>1. A merge is a hypothesis, not a fact.</strong> Two names are compared, never collapsed. The
link between them is a graded edge that carries its evidence and can be split again by
lowering a threshold — the reasoning is never deleted.</p>

<p><strong>2. Evidence is typed by diagnostic strength.</strong> Infrastructure, code and operator
artefacts can bear on identity; shared tradecraft cannot. This is the trap: two actors can
share tooling, binaries and C2 paths without being one organisation, so TTP overlap must
not, on its own, sustain an identity claim.</p>

<p><strong>3. Two axes, per claim.</strong> <em>Confidence</em> — how good the sourcing and reasoning are — is
kept separate from <em>probability</em> — how likely the outcome is. A claim can be very likely
at low confidence (a single source) or roughly even at high confidence (strong data,
genuine uncertainty). Conflating the two is the most common tradecraft error.</p>

<p><strong>4. Truth is external, and disagreement is a result.</strong> A binary judgement is admitted
only if independent sources actually assert it; a single source is not corroboration, and
a disagreement between sources is recorded as <em>contested</em>, not decided. An alias list is
weaker than an identity claim, and the method says so.</p>

<h2 id="from-method-to-tooling">From method to tooling</h2>

<p>Prose does not stop you skipping the hard part on a busy Tuesday; a program does. So each
rule above is encoded. The tools are small on purpose — the interesting thing is the rule,
not the line count. What they share is that they refuse to let a claim pass without its
evidence, its confidence, and a statement of what would change it. The
<a href="/engineering/">engineering page</a> walks through each one.</p>

<h2 id="what-this-does-not-prove">What this does not prove</h2>

<ul>
  <li>The checks are heuristics where the truth is not a schema; they are stated as such.</li>
  <li>External truth for a <em>specific</em> alias pair is scarce; a corpus that is mostly contested
is a finding, not a failure.</li>
  <li>The method is single-author. It is a standard, not a peer review.</li>
</ul>

<h2 id="the-numbers">The numbers</h2>

<p>The suite is public and MIT-licensed, and it runs in CI:
<a href="https://github.com/Falsifyer/Defensible-CTI">github.com/Falsifyer/Defensible-CTI</a>. I don’t copy test counts into
prose — a number in a blog post goes stale, and the whole point of the tooling is that the claim can
be checked at source. What the CI checks:</p>

<ul>
  <li>the reconciliation engine against an adjudicated corpus with a committed coverage baseline, with
STIX output validated against the reference library;</li>
  <li>the handling gate against FIRST TLP 2.0 and STIX 2.1, where the ground truth is a schema and not
an opinion;</li>
  <li>the hunting planner and the assessment ledger, whose <em>refusals</em> are the behaviour under test.</li>
</ul>

<hr />

<p><em>A judgement is finished only when it changes a decision — and when someone can see why.</em></p>

<h2 id="revision-history">Revision history</h2>

<table>
  <thead>
    <tr>
      <th>Version</th>
      <th>Date</th>
      <th>Change</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>1.0</td>
      <td>2026-10-09</td>
      <td>First release.</td>
    </tr>
    <tr>
      <td>1.1</td>
      <td>2026-10-28</td>
      <td>Handling corrected from the retired TLP 1.0 label to <code class="language-plaintext highlighter-rouge">TLP:CLEAR</code>; test counts replaced by the CI they come from. No assessed fact changed.</td>
    </tr>
  </tbody>
</table>]]></content><author><name>Martín Eliseo</name></author><category term="threat-intelligence" /><summary type="html"><![CDATA[How a source becomes a defensible judgement: two-axis grading, ACH, a computed ceiling and a corrections loop that leaves a trace.]]></summary></entry><entry><title type="html">Volt Typhoon: an actor profile built from open sources</title><link href="https://martineliseo.es/volt-typhoon-actor-profile/" rel="alternate" type="text/html" title="Volt Typhoon: an actor profile built from open sources" /><published>2026-10-07T00:00:00+00:00</published><updated>2026-10-07T00:00:00+00:00</updated><id>https://martineliseo.es/volt-typhoon-actor-profile</id><content type="html" xml:base="https://martineliseo.es/volt-typhoon-actor-profile/"><![CDATA[<table>
  <thead>
    <tr>
      <th> </th>
      <th> </th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td><strong>Product</strong></td>
      <td>Actor profile</td>
    </tr>
    <tr>
      <td><strong>Handling</strong></td>
      <td>TLP:CLEAR — open sources only</td>
    </tr>
    <tr>
      <td><strong>Version</strong></td>
      <td>2.4 (2026-10-28)</td>
    </tr>
    <tr>
      <td><strong>Review trigger</strong></td>
      <td>MITRE ATT&amp;CK or government update, a confirmed disruptive effect, or new edge-device exploitation</td>
    </tr>
    <tr>
      <td><strong>Next review</strong></td>
      <td>2027-04-30</td>
    </tr>
    <tr>
      <td><strong>Author</strong></td>
      <td>Martín Eliseo</td>
    </tr>
  </tbody>
</table>

<p><em>Every claim below carries a source and is graded on two independent axes. Where vendors disagree, I
say so.</em></p>

<h2 id="how-to-read-this">How to read this</h2>

<p>Two axes are graded separately, per claim — not per section:</p>

<ul>
  <li><strong>Confidence</strong> — how good the sourcing and the reasoning are: <strong>High</strong> (corroborated by two or more
independent authoritative sources), <strong>Moderate</strong> (one credible source, or corroboration with gaps),
<strong>Low</strong> (single-source, self-claimed, or heavily inferential).</li>
  <li><strong>Probability</strong> — how likely the assessed outcome is, stated only where the claim is estimative:
<em>nearly certain</em>, <em>very likely</em>, <em>likely</em>, <em>roughly even chance</em>, <em>unlikely</em>, <em>very unlikely</em>.</li>
</ul>

<p>They are not the same thing. A claim can be <em>very likely</em> at <em>low confidence</em> (a single source) or
<em>roughly even</em> at <em>high confidence</em> (strong data, genuine uncertainty). Purely factual claims are
graded for confidence only and carry <em>n/a</em> for probability. <strong>Unverified</strong> means it could not be
substantiated, and is recorded so it is not repeated as fact.</p>

<h2 id="bottom-line">Bottom line</h2>

<p>Volt Typhoon is a People’s Republic of China (PRC) state-sponsored intrusion set, active since at
least 2021, that targets critical infrastructure — above all <strong>communications, energy, transportation
and water/wastewater</strong> — in the continental United States, its territories (including Guam) and
Western-aligned countries. Its defining trait is not a tool but the absence of one: it prefers stolen
credentials, living-off-the-land (LOTL) binaries, hands-on-keyboard activity and web shells over
custom malware, and it hides behind chains of compromised SOHO routers and edge devices. U.S.
agencies assess the activity as <strong>pre-positioning</strong> for disruptive or destructive operations against
operational technology (OT) in the event of a crisis or conflict, not as espionage alone. Since 2025
the emphasis has moved from access to preparation: OT-focused reporting places a closely overlapping
cluster inside engineering workstations and cellular gateways — Stage 2 of the ICS Cyber Kill Chain —
with initial access increasingly brokered by a separate access cluster, SYLVANITE.</p>

<h2 id="1-identity-and-naming">1. Identity and naming</h2>

<p>The same activity is tracked under several designations:</p>

<table>
  <thead>
    <tr>
      <th>Designation</th>
      <th>Attributed to</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>Volt Typhoon</td>
      <td>Microsoft</td>
    </tr>
    <tr>
      <td>BRONZE SILHOUETTE</td>
      <td>Secureworks</td>
    </tr>
    <tr>
      <td>Vanguard Panda</td>
      <td>CrowdStrike</td>
    </tr>
    <tr>
      <td>DEV-0391, Storm-0391</td>
      <td>Microsoft (initial designations)</td>
    </tr>
    <tr>
      <td>UNC3236</td>
      <td>Mandiant / Google</td>
    </tr>
    <tr>
      <td>Voltzite</td>
      <td>Dragos</td>
    </tr>
    <tr>
      <td>Insidious Taurus</td>
      <td>Palo Alto Networks Unit 42</td>
    </tr>
    <tr>
      <td>Redfly</td>
      <td>Symantec / Gen Digital</td>
    </tr>
    <tr>
      <td>DazedToad</td>
      <td>MITRE ATT&amp;CK</td>
    </tr>
    <tr>
      <td>G1017</td>
      <td>MITRE ATT&amp;CK</td>
    </tr>
  </tbody>
</table>

<p>MITRE records these as <strong>Associated Groups</strong> [3], which is explicitly weaker than “same actor”.
CISA lists Vanguard Panda, BRONZE SILHOUETTE, Dev-0391, UNC3236, Voltzite and Insidious Taurus as
aliases in its advisory [1].</p>

<p><strong>Caveat — low confidence.</strong> Not every tracking name is asserted to be the same actor. Dragos’s
VOLTZITE is described as sharing “extensive technical overlaps” with Volt Typhoon rather than being
declared identical [7], and it is Dragos’s analyst designation for the OT-facing activity. Read the
OT reporting as an <strong>overlapping cluster</strong>, not a single verified entity [3][7]. The same caution
applies one layer down: the access cluster that feeds this actor is itself a set of names, not a
name (§3).</p>

<h2 id="2-assessment-ledger">2. Assessment ledger<!-- claims:begin --></h2>

<table>
  <thead>
    <tr>
      <th>Assessment</th>
      <th>Confidence</th>
      <th>Probability</th>
      <th>Basis</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>PRC state-sponsored, active since at least 2021</td>
      <td>Moderate</td>
      <td>n/a</td>
      <td>[1][3]</td>
    </tr>
    <tr>
      <td>Targets communications, energy, transportation and water/wastewater in the US and its territories, including Guam</td>
      <td>Moderate</td>
      <td>n/a</td>
      <td>[1]</td>
    </tr>
    <tr>
      <td>Assessed as pre-positioning for disruptive or destructive attacks against OT</td>
      <td>Moderate</td>
      <td>likely</td>
      <td>[1][3]</td>
    </tr>
    <tr>
      <td>Living-off-the-land tradecraft: stolen credentials, <code class="language-plaintext highlighter-rouge">wmic</code>/<code class="language-plaintext highlighter-rouge">ntdsutil</code>/<code class="language-plaintext highlighter-rouge">netsh</code>/PowerShell, minimal custom malware</td>
      <td>Moderate</td>
      <td>n/a</td>
      <td>[1][3]</td>
    </tr>
    <tr>
      <td>Uses compromised SOHO routers (KV Botnet and successors) to proxy C2 and obscure origin</td>
      <td>High</td>
      <td>n/a</td>
      <td>[1][6][8]</td>
    </tr>
    <tr>
      <td>Exploited the Versa Director zero-day CVE-2024-39717 with the VersaMem web shell</td>
      <td>Moderate</td>
      <td>n/a</td>
      <td>[3][9]</td>
    </tr>
    <tr>
      <td>Deployed the SockDetour backdoor in at least one intrusion</td>
      <td>Moderate</td>
      <td>n/a</td>
      <td>[10]</td>
    </tr>
    <tr>
      <td>Initial access increasingly brokered by SYLVANITE and handed to Volt Typhoon/VOLTZITE</td>
      <td>Moderate</td>
      <td>likely</td>
      <td>[7][12]</td>
    </tr>
    <tr>
      <td>Reached Stage 2 of the ICS Cyber Kill Chain in 2025 (engineering workstations, cellular gateways)</td>
      <td>Moderate</td>
      <td>n/a</td>
      <td>[7]</td>
    </tr>
    <tr>
      <td>Operated by the PLA Cyberspace Force</td>
      <td>Low</td>
      <td>n/a</td>
      <td>[11]</td>
    </tr>
    <tr>
      <td>Dragos’s VOLTZITE is the same actor as Volt Typhoon</td>
      <td>Low</td>
      <td>roughly even chance</td>
      <td>[3][7] · <a href="/ach-voltzite-volt-typhoon/">ACH</a></td>
    </tr>
  </tbody>
</table>

<!-- claims:end -->

<h2 id="3-tradecraft">3. Tradecraft</h2>

<p><strong>Living off the land, by design.</strong> The actor’s signature is doing almost everything with tools that
are already on the host: <code class="language-plaintext highlighter-rouge">wmic</code>, <code class="language-plaintext highlighter-rouge">ntdsutil</code>, <code class="language-plaintext highlighter-rouge">netsh</code> and PowerShell, plus built-in remote-access
features, all under valid — usually stolen — accounts. The effect is that its traffic and its
process tree look like an administrator’s. CISA’s advisory documents the command lines and publishes
detection signatures precisely because there is no malware to fingerprint [1][3].</p>

<p><strong>Infrastructure obfuscation and proxying.</strong> Rather than host its own command-and-control, it
builds chains of compromised small office/home office (SOHO) routers, mostly end-of-life Cisco and
Netgear devices, and proxies through them so that its traffic originates from U.S. IP addresses. The
FBI named the best-known instance the <strong>KV Botnet</strong> [6]; vendors and OT reporting have observed
successor botnets such as the JDY botnet after the takedown [7][8].</p>

<p><strong>Initial access through edge devices.</strong> Volt Typhoon reaches victims by exploiting internet-facing
edge infrastructure — VPN gateways, firewalls and SD-WAN management — rather than user inboxes. In
mid-2024 it exploited a zero-day in Versa Director, CVE-2024-39717, tracked by MITRE as campaign
C0039, dropping a custom web shell (VersaMem) whose purpose was credential capture from managed
service providers and ISPs, to pivot into their downstream customers [3][9].</p>

<p><strong>The access layer is tracked separately, and by one vendor.</strong> <strong>SYLVANITE</strong> is Dragos’s designation
for an initial-access group active since 2023 that weaponises edge-device proofs of concept, harvests
credentials and hands established footholds to Stage 2 actors. Dragos states that it has directly
observed SYLVANITE transferring access to VOLTZITE, with the handoff possible within days [12]. Two
cautions belong here.</p>

<p>First, <strong>MITRE does not record SYLVANITE as a group.</strong> ATT&amp;CK v20 — the version current when this
was checked — carries 193 enterprise groups and none is SYLVANITE, and G1017 does not mention it. What
MITRE records is <em>campaign</em> C0029, Cutting Edge — Ivanti Connect Secure zero-day exploitation from
December 2023 to February 2024, attributed variously to UNC5221/UTA0178 and UNC5325 [13]. A campaign
and a group are different objects, and citing one for the other is how a broker acquires a name it was
never given.</p>

<p>Second, <strong>SYLVANITE is not a single name either.</strong> Dragos lists it as overlapping UNC5221, UNC5174,
UNC5291, UNC3236, HOUKEN, Red Dev 61, CL-STA-0048 and UTA0178, and assesses those as <em>multiple
entities under the same overarching direction</em> [12]. Note what that list contains: UNC3236 is, in
ATT&amp;CK, an alias of Volt Typhoon [3]. Read flatly — the convention most platforms use — UNC3236 is
Volt Typhoon, and UNC3236 overlaps SYLVANITE, so SYLVANITE <em>is</em> Volt Typhoon, and the broker
disappears from the graph. No vendor asserts that. Overlap is not identity, and shared tooling is not
a shared operator: the honest unit here is an <strong>access ecosystem</strong>, not a broker.</p>

<p><strong>Persistence and credential theft.</strong> Persistence leans on web shells and valid accounts, with at
least one documented use of <strong>SockDetour</strong>, a custom backdoor designed to survive the removal of a
primary access method [10]. The collection priority is credentials and the material needed to
understand and reach the target’s operational environment.</p>

<h2 id="4-targeting">4. Targeting</h2>

<p>CISA’s advisory is explicit about sectors — <strong>Communications, Energy, Transportation Systems, and
Water and Wastewater Systems</strong> — across the continental and non-continental United States, including
Guam [1]. Microsoft’s tracking adds manufacturing, utilities, construction, maritime, government,
information technology, education, defense and media [4]. OT-focused reporting adds the operational
data itself: geographic information system (GIS) data, OT network diagrams and OT operating
instructions exfiltrated from ICS organisations [7][8]. In 2025, Dragos reported activity reaching
U.S. midstream pipeline operations through compromised Sierra Wireless AirLink cellular gateways,
and electric and oil-and-gas environments [7].</p>

<h2 id="5-from-access-to-pre-positioning-20252026">5. From access to pre-positioning (2025–2026)</h2>

<p>This is the assessment that makes Volt Typhoon different from a data-theft crew, and it is the one
to hold carefully. The U.S. government’s stated view is that the PRC is positioning itself to
launch destructive attacks that would threaten physical safety and impede military readiness in a
crisis or conflict [1] — an assessment about <strong>intent and preparation</strong>, not a report of observed
destruction.</p>

<p>What changed in 2025 is that OT reporting moved the activity further along that path. Dragos placed
VOLTZITE at <strong>Stage 2 of the ICS Cyber Kill Chain</strong>: manipulating engineering workstation software
to extract configuration files and alarm data, and specifically investigating what operational
conditions would trigger process shutdowns — reconnaissance of <em>how to disrupt</em>, not disruption
itself [7]. Initial access is increasingly brokered: SYLVANITE weaponises edge-device vulnerabilities
and hands footholds over for the deeper OT intrusion [7][12]. No publicly confirmed disruptive effect
has been observed. What is <strong>corroborated</strong> here is only that the actor has the access and is
collecting the data; the disruptive end state remains an assessed intent [1][7].</p>

<h2 id="6-disruption-and-current-status">6. Disruption and current status</h2>

<ul>
  <li><strong>KV Botnet disrupted.</strong> Following a December 2023 court-authorised operation, the FBI announced on
31 January 2024 that it had remotely removed the KV Botnet malware from hundreds of U.S.-based
SOHO routers and severed their command-and-control [6].</li>
  <li><strong>Temporary, and replaced.</strong> The FBI noted the remediation was temporary — a router reboot could
reverse it — and that the devices remained vulnerable [6]. Dragos observed the botnet being
rebuilt and successor infrastructure in use afterwards [7][8].</li>
  <li><strong>Not deterred.</strong> Activity continued through 2025 and, by OT reporting, deepened [7].</li>
  <li>Assessment: <strong>infrastructure degraded, intent unchanged</strong>. Takedowns cost the actor proxies, not
access to its objectives [6][7].</li>
</ul>

<h2 id="7-what-this-means-for-defenders">7. What this means for defenders</h2>

<p>Ordered by how directly they attack this actor’s playbook:</p>

<ol>
  <li><strong>Inventory and patch the edge.</strong> VPN gateways, firewalls and SD-WAN managers are the front door.
Known-exploited CVEs here, and end-of-life devices, are the single highest-value fix [9].</li>
  <li><strong>Assume stolen credentials are in play.</strong> Phishing-resistant MFA, credential hygiene and
privilege tiering — the actor’s whole method depends on valid accounts that look legitimate [1].</li>
  <li><strong>Log the identity and remote-access planes.</strong> Authentication, VPN and RDP events, plus process
telemetry (EDR/Sysmon), because the malicious activity is native behaviour [1].</li>
  <li><strong>Hunt the LOTL command lines.</strong> <code class="language-plaintext highlighter-rouge">ntdsutil</code> credential extraction, <code class="language-plaintext highlighter-rouge">wmic</code>/<code class="language-plaintext highlighter-rouge">netsh</code> used by
unexpected parents, and PowerShell that does not match administrative baselines [1][3].</li>
  <li><strong>Watch egress to SOHO and edge ranges.</strong> Encrypted sessions to compromised consumer routers, and
the non-standard TCP-then-HTTPS pattern MITRE documents for C0039, are high-signal [3][9].</li>
  <li><strong>Segment and monitor OT.</strong> Engineering-workstation configuration and alarm data, and cellular
gateways, are the current collection targets — monitor reads, not just writes [7].</li>
  <li><strong>Treat “no alerts” as “no evidence”.</strong> For an actor whose method is native tools and valid
accounts, absence of detections is not absence of the actor [1][3].</li>
</ol>

<h2 id="8-detection-opportunities">8. Detection opportunities</h2>

<p>What to look for, not a validated ruleset (consistent with the rule-provenance policy I use elsewhere):</p>

<ul>
  <li><code class="language-plaintext highlighter-rouge">ntdsutil</code> invoking <code class="language-plaintext highlighter-rouge">ac i ntds</code> to create an IFM (Install From Media) copy — credential dumping that
needs no external tool.</li>
  <li><code class="language-plaintext highlighter-rouge">wmic</code>, <code class="language-plaintext highlighter-rouge">netsh</code> or PowerShell spawned by service, web-server or backup processes rather than an
interactive admin session.</li>
  <li>A new web shell or unexpected file on an SD-WAN manager, VPN appliance or firewall.</li>
  <li>Long-lived encrypted sessions from a host to consumer/SOHO router address space, and the
non-standard TCP handshake followed by HTTPS command-and-control from campaign C0039 [3][9].</li>
  <li>On OT networks: reads of GIS data, OT network diagrams, alarm configuration or engineering
workstation configuration files by accounts outside the engineering baseline [7].</li>
</ul>

<h2 id="9-intelligence-gaps-and-what-i-could-not-verify">9. Intelligence gaps and what I could not verify</h2>

<ul>
  <li>Whether VOLTZITE (Dragos) and Volt Typhoon (Microsoft and the U.S. government) are one actor or an
overlapping ecosystem; the naming is explicitly “technical overlap”, not identity [3][7].</li>
  <li>Whether the eight names Dragos lists as overlapping SYLVANITE are one entity. Dragos assesses
multiple entities under one overarching direction, which is not the same as a single broker, and no
public source resolves it [12].</li>
  <li>Unit-level attribution — the claim that the group is run by the PLA Cyberspace Force is public
reporting, not a government confirmation, and is treated as low confidence [11].</li>
  <li>Whether the pre-positioning has translated into any confirmed disruptive or destructive effect.
None is public; the disruptive end state is an assessed intent [1][7].</li>
  <li>Victim enumeration is inherently incomplete — CISA’s own framing is that what has been found is
likely “the tip of the iceberg” [1].</li>
  <li>Nothing here is validated against private telemetry, and I did not analyse malware samples for this
profile. This is an open-source assessment, and it is bounded by that.</li>
</ul>

<h2 id="10-outlook--what-would-change-this-judgement">10. Outlook — what would change this judgement</h2>

<ul>
  <li>A <strong>MITRE ATT&amp;CK or government update</strong> that merges or separates VOLTZITE and Volt Typhoon, or
reassigns the group/campaign IDs.</li>
  <li>Any <strong>publicly confirmed disruptive or destructive effect</strong>; that would move the pre-positioning
judgement from assessed intent to observed action.</li>
  <li>New <strong>edge-device exploitation</strong> (a fresh known-exploited CVE) or a new access-broker cluster.</li>
  <li>A change in the <strong>SOHO botnet</strong> layer — successor infrastructure, or a takedown that holds.</li>
</ul>

<h2 id="sources">Sources</h2>

<ol>
  <li>CISA, NSA, FBI and international partners — <strong>Joint Cybersecurity Advisory AA24-038A: PRC
State-Sponsored Actors Compromise and Maintain Persistent Access to U.S. Critical Infrastructure</strong>
(7 Feb 2024). <a href="https://www.cisa.gov/news-events/cybersecurity-advisories/aa24-038a">https://www.cisa.gov/news-events/cybersecurity-advisories/aa24-038a</a></li>
  <li>CISA, NSA, FBI and partners — <strong>Joint Cybersecurity Advisory AA23-144A: PRC State-Sponsored
Cyber Actor Living off the Land to Evade Detection</strong> (24 May 2023).
<a href="https://www.cisa.gov/news-events/cybersecurity-advisories/aa23-144a">https://www.cisa.gov/news-events/cybersecurity-advisories/aa23-144a</a></li>
  <li>MITRE ATT&amp;CK — <strong>Group G1017: Volt Typhoon</strong> (v2.0, last modified 31 Jul 2026).
<a href="https://attack.mitre.org/groups/G1017">https://attack.mitre.org/groups/G1017</a></li>
  <li>Microsoft Threat Intelligence — <strong>Volt Typhoon (VANGUARD PANDA)</strong>.
<a href="https://www.microsoft.com/en-us/security/security-insider/threat-landscape/volt-typhoon">https://www.microsoft.com/en-us/security/security-insider/threat-landscape/volt-typhoon</a></li>
  <li>CISA, NSA and partners — <strong>Identifying and Mitigating Living Off the Land Techniques</strong> (joint
guidance, Feb 2024). <a href="https://www.cisa.gov/resources-tools/resources/identifying-and-mitigating-living-land-techniques">https://www.cisa.gov/resources-tools/resources/identifying-and-mitigating-living-land-techniques</a></li>
  <li>U.S. Department of Justice — <strong>U.S. Government Disrupts Botnet PRC Used to Conceal Hacking of
Critical Infrastructure</strong> (31 Jan 2024).
<a href="https://www.justice.gov/opa/pr/us-government-disrupts-botnet-peoples-republic-china-used-conceal-hacking-critical">https://www.justice.gov/opa/pr/us-government-disrupts-botnet-peoples-republic-china-used-conceal-hacking-critical</a></li>
  <li>Dragos — <strong>OT/ICS Cybersecurity Report 2026: Year in Review</strong> (17 Feb 2026).
<a href="https://www.dragos.com/resources/press-release/dragos-2026-year-in-review-new-ot-threats-ransomware">https://www.dragos.com/resources/press-release/dragos-2026-year-in-review-new-ot-threats-ransomware</a></li>
  <li>Dragos — <strong>OT/ICS Cybersecurity Report 2025: Year in Review</strong> (25 Feb 2025).
<a href="https://www.dragos.com/resources/press-release/dragos-reports-ot-ics-cyber-threats-escalate-amid-geopolitical-conflicts-and-increasing-ransomware-attacks">https://www.dragos.com/resources/press-release/dragos-reports-ot-ics-cyber-threats-escalate-amid-geopolitical-conflicts-and-increasing-ransomware-attacks</a></li>
  <li>MITRE ATT&amp;CK — <strong>Campaign C0039: Versa Director Zero Day Exploitation</strong> (June–August 2024), and
Lumen Black Lotus Labs’ exploitation report.
<a href="https://attack.mitre.org/campaigns/C0039">https://attack.mitre.org/campaigns/C0039</a></li>
  <li>Palo Alto Networks Unit 42 — <strong>Threat Brief: Attacks on Critical Infrastructure Attributed to
Insidious Taurus (Volt Typhoon)</strong> (Feb 2024). <a href="https://unit42.paloaltonetworks.com/volt-typhoon-threat-brief/">https://unit42.paloaltonetworks.com/volt-typhoon-threat-brief/</a></li>
  <li>Public reporting on unit-level attribution of the group (secondary source; treated as low
confidence).</li>
  <li>Dragos — <strong>SYLVANITE</strong> threat group page (accessed 28 Oct 2026). Records the group since 2023,
the observed handoff to VOLTZITE, and the overlap list.
<a href="https://www.dragos.com/threat/sylvanite/">https://www.dragos.com/threat/sylvanite/</a></li>
  <li>MITRE ATT&amp;CK — <strong>Campaign C0029: Cutting Edge</strong> (Ivanti Connect Secure exploitation,
December 2023 – February 2024; last modified 31 Jul 2026).
<a href="https://attack.mitre.org/campaigns/C0029">https://attack.mitre.org/campaigns/C0029</a></li>
</ol>

<hr />

<p><em>All analysis is open-source. Views are my own.</em></p>

<h2 id="revision-history">Revision history</h2>

<table>
  <thead>
    <tr>
      <th>Version</th>
      <th>Date</th>
      <th>Change</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>1.0</td>
      <td>2026-10-07</td>
      <td>First release.</td>
    </tr>
    <tr>
      <td>2.0</td>
      <td>2026-10-09</td>
      <td>Split confidence and probability into two axes; added the product header, the outlook and this history. No assessed fact changed.</td>
    </tr>
    <tr>
      <td>2.1</td>
      <td>2026-10-28</td>
      <td>The VOLTZITE identity row now carries the ACH’s judgement (<code class="language-plaintext highlighter-rouge">unlikely</code>, ~20–45%) instead of <code class="language-plaintext highlighter-rouge">roughly even chance</code>, and points to the product that tests it. Handling corrected from the retired TLP 1.0 label to <code class="language-plaintext highlighter-rouge">TLP:CLEAR</code>, now checked in CI. No other assessed fact changed.</td>
    </tr>
    <tr>
      <td>2.2</td>
      <td>2026-10-28</td>
      <td>Corrected the SYLVANITE attribution: MITRE records campaign C0029, not a group, and G1017 does not mention SYLVANITE. Added the overlap list — eight names, assessed by Dragos as multiple entities under one direction — including UNC3236, an ATT&amp;CK alias of Volt Typhoon, and the reason overlap must not be read as identity. Ledger basis for the broker row moved from [3][7] to [7][12].</td>
    </tr>
    <tr>
      <td>2.3</td>
      <td>2026-10-28</td>
      <td>The identity row follows ACH v2.0: <code class="language-plaintext highlighter-rouge">roughly even chance</code> instead of <code class="language-plaintext highlighter-rouge">unlikely</code>, because v2 counts the naming observation once instead of three times. The probability changed; the confidence did not.</td>
    </tr>
    <tr>
      <td>2.4</td>
      <td>2026-10-28</td>
      <td>Grades now computed from the claim registry. Five <code class="language-plaintext highlighter-rouge">High</code> became <code class="language-plaintext highlighter-rouge">Moderate</code>: MITRE G1017 derives from the CISA advisory, so it is not a second voice, and two ATT&amp;CK objects are one publisher. The facts are unchanged; the confidence was overstated.</td>
    </tr>
  </tbody>
</table>]]></content><author><name>Martín Eliseo</name></author><category term="threat-intelligence" /><summary type="html"><![CDATA[A graded profile of the PRC-linked actor: living-off-the-land tradecraft, SOHO proxying and pre-positioning for OT disruption.]]></summary></entry><entry><title type="html">Scattered Spider: an actor profile built from open sources</title><link href="https://martineliseo.es/scattered-spider-actor-profile/" rel="alternate" type="text/html" title="Scattered Spider: an actor profile built from open sources" /><published>2026-10-06T00:00:00+00:00</published><updated>2026-10-06T00:00:00+00:00</updated><id>https://martineliseo.es/scattered-spider-actor-profile</id><content type="html" xml:base="https://martineliseo.es/scattered-spider-actor-profile/"><![CDATA[<table>
  <thead>
    <tr>
      <th> </th>
      <th> </th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td><strong>Product</strong></td>
      <td>Actor profile</td>
    </tr>
    <tr>
      <td><strong>Handling</strong></td>
      <td>TLP:CLEAR — open sources only</td>
    </tr>
    <tr>
      <td><strong>Version</strong></td>
      <td>2.2 (2026-10-28)</td>
    </tr>
    <tr>
      <td><strong>Review trigger</strong></td>
      <td>New government advisory, MITRE ATT&amp;CK update, vendor re-assessment, or law-enforcement action</td>
    </tr>
    <tr>
      <td><strong>Next review</strong></td>
      <td>2027-04-30</td>
    </tr>
    <tr>
      <td><strong>Author</strong></td>
      <td>Martín Eliseo</td>
    </tr>
  </tbody>
</table>

<p><em>Every claim below carries a source and is graded on two independent axes. Where vendors disagree, I
say so.</em></p>

<h2 id="how-to-read-this">How to read this</h2>

<p>Two axes are graded separately, per claim — not per section:</p>

<ul>
  <li><strong>Confidence</strong> — how good the sourcing and the reasoning are: <strong>High</strong> (corroborated by two or more
independent authoritative sources), <strong>Moderate</strong> (one credible source, or corroboration with gaps),
<strong>Low</strong> (single-source, self-claimed, or heavily inferential).</li>
  <li><strong>Probability</strong> — how likely the assessed outcome is, stated only where the claim is estimative:
<em>nearly certain</em>, <em>very likely</em>, <em>likely</em>, <em>roughly even chance</em>, <em>unlikely</em>, <em>very unlikely</em>.</li>
</ul>

<p>They are not the same thing. A claim can be <em>very likely</em> at <em>low confidence</em> (a single source) or
<em>roughly even</em> at <em>high confidence</em> (strong data, genuine uncertainty). Purely factual claims are
graded for confidence only and carry <em>n/a</em> for probability. <strong>Unverified</strong> means it could not be
substantiated, and is recorded so it is not repeated as fact.</p>

<h2 id="bottom-line">Bottom line</h2>

<p>Scattered Spider is a native-English-speaking, financially motivated intrusion set active since at
least 2022 whose primary attack surface is <strong>people and identity, not software</strong>. Its signature is
impersonating IT and help-desk staff to reset passwords and move MFA tokens, then abusing legitimate
remote-monitoring tools and cloud identity to hold an enterprise from the inside. Since 2024 it has
added hypervisor-level (VMware vSphere/ESXi) operations and ransomware to a data-extortion playbook,
and in 2025 a self-branded coalition — “Scattered Lapsus$ Hunters” — ran a large voice-phishing and
OAuth-theft campaign against SaaS (Salesforce) tenants. Law-enforcement arrests in 2025 slowed the
activity without stopping it.</p>

<h2 id="1-identity-and-naming">1. Identity and naming</h2>

<p>The same activity is tracked under several designations:</p>

<table>
  <thead>
    <tr>
      <th>Designation</th>
      <th>Attributed to</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>Scattered Spider (early phase: Roasted 0ktapus)</td>
      <td>CrowdStrike</td>
    </tr>
    <tr>
      <td>UNC3944</td>
      <td>Mandiant / Google Threat Intelligence</td>
    </tr>
    <tr>
      <td>Octo Tempest, Storm-0875</td>
      <td>Microsoft</td>
    </tr>
    <tr>
      <td>Muddled Libra</td>
      <td>Palo Alto Networks Unit 42</td>
    </tr>
    <tr>
      <td>0ktapus</td>
      <td>Group-IB</td>
    </tr>
    <tr>
      <td>Scatter Swine</td>
      <td>Okta</td>
    </tr>
    <tr>
      <td>G1015</td>
      <td>MITRE ATT&amp;CK</td>
    </tr>
  </tbody>
</table>

<p>CISA lists “UNC3944, Scatter Swine, Oktapus, Octo Tempest, Storm-0875, and Muddled Libra” as aliases
of Scattered Spider [1]; MITRE records them as <strong>Associated Groups</strong> [2], which is explicitly weaker
than “same actor”.</p>

<p><strong>Caveat — low confidence.</strong> These designations are not universally treated as co-referential. Public
analysis notes that the shared 0ktapus phishing kit was reused by multiple actors, and some vendor
assessments go as far as treating Muddled Libra as a distinct cluster despite overlapping tradecraft.
Read the aliases as an <strong>overlapping cluster, not a verified single identity</strong> [2][9].</p>

<h2 id="2-assessment-ledger">2. Assessment ledger<!-- claims:begin --></h2>

<table>
  <thead>
    <tr>
      <th>Assessment</th>
      <th>Confidence</th>
      <th>Probability</th>
      <th>Basis</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>Native-English-speaking, active since at least 2022</td>
      <td>Moderate</td>
      <td>n/a</td>
      <td>[2]</td>
    </tr>
    <tr>
      <td>Initial access via help-desk vishing and MFA bypass (push bombing, SIM swap)</td>
      <td>Moderate</td>
      <td>n/a</td>
      <td>[1][2]</td>
    </tr>
    <tr>
      <td>Abuse of legitimate RMM tools (AnyDesk, TeamViewer, ScreenConnect, Splashtop, Tailscale, Teleport.sh)</td>
      <td>Moderate</td>
      <td>n/a</td>
      <td>[1]</td>
    </tr>
    <tr>
      <td>Registers its own MFA tokens and historically adds a federated IdP to the SSO tenant</td>
      <td>Moderate</td>
      <td>n/a</td>
      <td>[1]</td>
    </tr>
    <tr>
      <td>Hypervisor (vCenter/ESXi) operations</td>
      <td>High</td>
      <td>n/a</td>
      <td>[2][3]</td>
    </tr>
    <tr>
      <td>Ransomware: BlackCat, and DragonForce from 2025</td>
      <td>Moderate</td>
      <td>n/a</td>
      <td>[1][2]</td>
    </tr>
    <tr>
      <td>Sector wave targeting (retail, insurance, airlines)</td>
      <td>Moderate</td>
      <td>likely</td>
      <td>[3][4]</td>
    </tr>
    <tr>
      <td>2025 arrests slowed the group</td>
      <td>Moderate</td>
      <td>likely</td>
      <td>[5]</td>
    </tr>
    <tr>
      <td>Scattered Lapsus$ Hunters is Scattered Spider plus Lapsus$ plus ShinyHunters</td>
      <td>Low</td>
      <td>roughly even chance</td>
      <td>[6][7][8]</td>
    </tr>
    <tr>
      <td>The ~1 billion-record Salesforce claim</td>
      <td>Low</td>
      <td>unlikely</td>
      <td>[6][7]</td>
    </tr>
  </tbody>
</table>

<!-- claims:end -->

<h2 id="3-tradecraft">3. Tradecraft</h2>

<p><strong>Initial access — the help desk is the perimeter.</strong> Voice phishing (vishing) impersonating IT or
help-desk staff — or employees impersonating staff — to get passwords reset and MFA tokens
transferred, frequently across several calls (“layered social engineering”); MFA push bombing (MFA
fatigue); SIM swap to intercept one-time codes; and purchased credentials from criminal marketplaces,
including abuse of contracted third-party IT help desks. Lookalike domains mirror the target
(<code class="language-plaintext highlighter-rouge">&lt;target&gt;-helpdesk.com</code>, <code class="language-plaintext highlighter-rouge">&lt;target&gt;-okta.com</code>, <code class="language-plaintext highlighter-rouge">&lt;target&gt;-servicedesk.com</code>) [1][2].</p>

<p><strong>Identity persistence and privilege.</strong> The actor registers its own MFA tokens and, historically, adds
a federated identity provider to the SSO tenant and enables automatic account linking — allowing
sign-in even after a password reset. It also modifies conditional-access policies and creates new
identities to survive response [1][2].</p>

<p><strong>Cloud and hypervisor.</strong> It enumerates and abuses Okta, AWS, and Microsoft Entra ID, uses AWS Systems
Manager Inventory for discovery, and creates cloud instances for lateral movement and staging [1][2].
The 2025 pivot is notable: compromise of vSphere (vCenter Server), reboot into single-user mode,
deployment of Teleport for covert access, mounting domain-controller disks via “orphaned” VMs,
disabling backups, and running ransomware <strong>from the ESXi shell</strong> — which renders in-guest security
tooling ineffective [3].</p>

<p><strong>Discovery, collection, exfiltration.</strong> It searches SharePoint, code repositories, credential-storage
documents, VPN/MFA enrollment guides and new-hire guides; reads Slack, Teams and Exchange for
conversations about the intrusion and the response; and — reported repeatedly — joins incident-response
calls to learn how defenders are hunting it [1][2]. Exfiltration targets include MEGA, AWS S3, and
Snowflake for mass extraction [1].</p>

<p><strong>Impact.</strong> Dual extortion: data theft with the threat to leak it, plus encryption (BlackCat
historically; DragonForce in 2025, encrypting ESXi servers) [1][2].</p>

<h2 id="4-targeting">4. Targeting</h2>

<p>Initially CRM providers, business-process outsourcing, telecom and technology; from 2023 gaming,
hospitality, retail, MSP, manufacturing and financial [2]. In 2025, a concentrated wave against UK
retail and — per Google — US retail, airline and insurance [3][4].</p>

<h2 id="5-the-scattered-lapsus-hunters-branding-2025--low-confidence">5. The “Scattered Lapsus$ Hunters” branding (2025) — low confidence</h2>

<p>A group self-branded <strong>Scattered Lapsus$ Hunters</strong> (also “Trinity of Chaos”), claiming to combine
Scattered Spider, Lapsus$ and ShinyHunters, ran a large campaign against Salesforce customers using
voice phishing and malicious OAuth applications — including abuse of Salesloft Drift tokens — with
claims reaching approximately one billion records across dozens of named organisations. Google tracks
related activity as UNC6040/UNC6395 and the FBI issued an advisory. Attribution rests largely on
<strong>actor self-claims plus TTP overlap</strong>; analysts caution the relationship may be a shared criminal
community rather than a single merged entity, and Salesforce disputes that its platform was
compromised. Treat this as context on a brand, not a confirmed organisational structure [6][7][8].</p>

<h2 id="6-disruption-and-current-status">6. Disruption and current status</h2>

<p>Arrests and guilty pleas in 2024–2026 prompted reporting in mid-2025 that UNC3944 activity had “gone
quiet”, followed by continued or rebranded activity. Assessment: <strong>degraded, not dismantled</strong> [5].</p>

<h2 id="7-what-this-means-for-defenders">7. What this means for defenders</h2>

<p>Adapted from CISA’s mitigations [1], ordered by how directly they attack this actor’s playbook:</p>

<ol>
  <li><strong>Phishing-resistant MFA (FIDO2/WebAuthn).</strong> This single control removes push bombing and SIM swap
from the board.</li>
  <li><strong>Help-desk verification.</strong> Out-of-band identity checks for password resets, no MFA-token transfer
over the phone, and alerting on reset-then-reenrolment bursts.</li>
  <li><strong>Control remote access tools.</strong> Allowlist execution and alert on portable RMM binaries.</li>
  <li><strong>Harden the hypervisor.</strong> No direct ESXi shell access, isolated/immutable backups, and monitoring
of vCenter SSO and new admin roles.</li>
  <li><strong>Watch the identity control plane.</strong> New federated IdPs, new trusted locations, and new MFA device
registrations are high-signal events here.</li>
  <li><strong>Offline, tested backups.</strong> The last line, and the one this actor specifically targets.</li>
</ol>

<h2 id="8-detection-opportunities">8. Detection opportunities</h2>

<p>What to look for, not a validated ruleset (consistent with the rule-provenance policy I use elsewhere):</p>

<ul>
  <li>A help-desk password reset followed within a short window by MFA re-registration for the same identity.</li>
  <li>A new federated identity provider or trusted location added to the IdP tenant.</li>
  <li>Portable RMM binaries running from user-writable paths, and their network beaconing.</li>
  <li>On vSphere: changes to the <code class="language-plaintext highlighter-rouge">ESX Admins</code> group, vCenter single-user-mode reboots, and Teleport
service units.</li>
</ul>

<h2 id="9-intelligence-gaps-and-what-i-could-not-verify">9. Intelligence gaps and what I could not verify</h2>

<ul>
  <li>The exact structure and membership of “Scattered Lapsus$ Hunters” — one entity, or a brand over a
community.</li>
  <li>Victim counts (actor claims, disputed by at least one platform vendor).</li>
  <li>The precise boundary between Scattered Spider, UNC3944 and Muddled Libra remains contested.</li>
  <li>Nothing here is validated against private telemetry, and I did not analyse malware samples for this
profile. This is an open-source assessment, and it is bounded by that.</li>
</ul>

<h2 id="10-outlook--what-would-change-this-judgement">10. Outlook — what would change this judgement</h2>

<p>The confidence above is a snapshot; these events would move it:</p>

<ul>
  <li>A <strong>new or revised advisory</strong> (CISA AA23-320A, or MITRE ATT&amp;CK group G1015) that merges the alias
cluster or splits it — the single strongest lever on the naming judgement.</li>
  <li>Vendor reporting that <strong>resolves Muddled Libra</strong> as the same actor or as a distinct cluster.</li>
  <li>Attribution of <strong>“Scattered Lapsus$ Hunters”</strong> beyond actor self-claims and TTP overlap.</li>
  <li>Evidence the <strong>ESXi/ransomware playbook</strong> has changed (a new encryptor, or new hypervisor tradecraft).</li>
</ul>

<h2 id="sources">Sources</h2>

<ol>
  <li>CISA, FBI, RCMP, ASD/ACSC, AFP, CCCS, NCSC-UK — <strong>Cybersecurity Advisory AA23-320A: Scattered
Spider</strong> (last revised 29 Jul 2025). <a href="https://www.cisa.gov/news-events/cybersecurity-advisories/aa23-320a">https://www.cisa.gov/news-events/cybersecurity-advisories/aa23-320a</a></li>
  <li>MITRE ATT&amp;CK — <strong>Group G1015: Scattered Spider</strong> (v3.0, last modified 31 Jul 2026).
<a href="https://attack.mitre.org/groups/G1015">https://attack.mitre.org/groups/G1015</a></li>
  <li>Google Threat Intelligence Group — <strong>Defending vSphere from UNC3944</strong> (summarised by Infosecurity
Magazine, 28 Jul 2025). <a href="https://cloud.google.com/blog/topics/threat-intelligence/defending-vsphere-from-unc3944">https://cloud.google.com/blog/topics/threat-intelligence/defending-vsphere-from-unc3944</a></li>
  <li>Google/GTIG via Computer Weekly — <em>Scattered Spider retail attacks spreading to US</em> (14 May 2025).
<a href="https://www.computerweekly.com/news/366623999/Scattered-Spider-retail-attacks-spreading-to-US-says-Google">https://www.computerweekly.com/news/366623999/Scattered-Spider-retail-attacks-spreading-to-US-says-Google</a></li>
  <li>Infosecurity Magazine — <em>Cybercriminals ‘Spooked’ After Scattered Spider Arrests</em> (31 Jul 2025).
<a href="https://www.infosecurity-magazine.com/news/cybercriminals-spooked-scattered">https://www.infosecurity-magazine.com/news/cybercriminals-spooked-scattered</a></li>
  <li>BleepingComputer — <em>ShinyHunters launches Salesforce data leak site to extort 39 victims</em> (3 Oct
2025). <a href="https://www.bleepingcomputer.com/news/security/shinyhunters-starts-leaking-data-stolen-in-salesforce-attacks">https://www.bleepingcomputer.com/news/security/shinyhunters-starts-leaking-data-stolen-in-salesforce-attacks</a></li>
  <li>SC Media — <em>New Scattered Lapsus$ Hunters escalates Salesforce extortion</em> (6 Oct 2025).
<a href="https://www.scworld.com/brief/new-scattered-lapsus-hunters-escalates-salesforce-extortion">https://www.scworld.com/brief/new-scattered-lapsus-hunters-escalates-salesforce-extortion</a></li>
  <li>Computer Weekly — <em>ShinyHunters Salesforce cyber attacks explained</em> (11 Aug 2025).
<a href="https://www.computerweekly.com/feature/ShinyHunters-Salesforce-cyber-attacks-explained-What-you-need-to-know">https://www.computerweekly.com/feature/ShinyHunters-Salesforce-cyber-attacks-explained-What-you-need-to-know</a></li>
  <li>Vendor tracking names and the non-equivalence caveat, as reported in public actor-profile
aggregations (secondary source; treated as low confidence).</li>
</ol>

<hr />

<p><em>All analysis is open-source. Views are my own.</em></p>

<h2 id="revision-history">Revision history</h2>

<table>
  <thead>
    <tr>
      <th>Version</th>
      <th>Date</th>
      <th>Change</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>1.0</td>
      <td>2026-10-06</td>
      <td>First release.</td>
    </tr>
    <tr>
      <td>2.0</td>
      <td>2026-10-09</td>
      <td>Split confidence and probability into two axes; added the product header, the outlook and this history. No assessed fact changed.</td>
    </tr>
    <tr>
      <td>2.1</td>
      <td>2026-10-28</td>
      <td>Handling corrected from the retired TLP 1.0 label to <code class="language-plaintext highlighter-rouge">TLP:CLEAR</code> (TLP 2.0), now checked in CI. No assessed fact changed.</td>
    </tr>
    <tr>
      <td>2.2</td>
      <td>2026-10-28</td>
      <td>Grades now computed from the claim registry. Five <code class="language-plaintext highlighter-rouge">High</code> became <code class="language-plaintext highlighter-rouge">Moderate</code>: MITRE G1015 derives from the CISA-led advisory, and several rows rested on one source. The facts are unchanged; the confidence was overstated.</td>
    </tr>
  </tbody>
</table>]]></content><author><name>Martín Eliseo</name></author><category term="threat-intelligence" /><summary type="html"><![CDATA[A graded actor profile: identity-first intrusion, RMM abuse, hypervisor operations and ransomware, with the vendor divergences left visible.]]></summary></entry></feed>