dogesec Logo
The Patch Nobody Applied: What the FBI Admitted About ShinyHunters

The Patch Nobody Applied: What the FBI Admitted About ShinyHunters

ShinyHunters said it used a zero-day on the FBI; the FBI blamed a contractor's missed patch. What 61 posts show, with a Sigma rule and STIX graphs.

Category: Research

tl;dr

  • On 22 September 2026 ShinyHunters defaced the FBI’s jobs site and claimed 2 to 3 TB of data on employees and applicants. Reporters saw a sample of about 5,000 employee records, with home addresses and intelligence roles.
  • The FBI blamed a contractor who failed to apply a security patch, and removed it on 5 October. Reuters’ sources named Accenture and Oracle PeopleSoft; the FBI has named neither.
  • The best-supported way in is a PeopleSoft flaw Oracle fixed on 10 June. The group says it got past firewall rules by encoding one letter of the URL. Nobody official has confirmed either.

“This site has been seized”

On Tuesday 22 September 2026, anyone visiting apply.fbijobs.gov, where people apply to work for the FBI, found a message instead of a form: “THIS SITE HAS BEEN SEIZED BY SHINYHUNTERS”. The group told BleepingComputer it had broken in on the Monday night with a new remote code execution flaw in Oracle PeopleSoft, moved on into FBI-managed AWS GovCloud (Amazon’s cloud region for US government workloads), and taken 2 to 3 TB on current and former employees and applicants from systems it called Criminal Justice, HR and MedLink. PeopleSoft is Oracle’s software for human resources, payroll and recruitment. It runs on a platform called PeopleTools and reaches browsers through WebLogic, Oracle’s Java application server.

The group’s stated motive was an FBI warning. On 15 May the FBI’s Internet Crime Complaint Center (IC3) had published advisory I-051526-PSA, saying ShinyHunters makes “real or exaggerated claims of access” and harasses victims, up to swatting: false emergency calls that send armed police to a victim’s home. The group wanted it retracted within a week, and told The Register the attack was “NOT financially motivated”. The FBI said only that it was “aware of claims regarding unauthorized activity affecting FBIJobs.gov”.

So the FBI had a breach claim from a group it had just warned the world exaggerates, and no public answer to the question that mattered: how did they get in?

61 posts and two stories

Obstracts collected 61 posts about the breach from 28 sources between 22 September and 9 October. Most repeat each other, and they tell two stories. For a fortnight the coverage followed the group’s “zero-day” framing. From 6 October most posts said “missed patch”, and seven named Accenture, which Reuters’ sources gave as the contractor.

A zero-day is a flaw with no fix yet; a missed patch is a fix that existed and wasn’t applied. They can’t both describe the same bug on the same night, which is why the primary sources matter:

Source What it saw What it lacks
BleepingComputer (22 Sept) The group’s own account: the defacement, the 2 to 3 TB, the “new” zero-day, GovCloud, the retraction demand Any check of those claims
Nextgov (23 Sept) The leaked sample: its fields and the roles in it How the data was taken
The FBI, via The Register (5 Oct) Arrests made How the group got in
The FBI, via Reuters and SecurityWeek (5 and 6 Oct) The victim’s account: a contractor’s missed patch on a third-party platform, the contractor removed The product, the vulnerability, the contractor’s name
Oracle (10 June) CVE-2026-35273 and its fix Any mention of the FBI
Google Threat Intelligence (11 June) The group’s zero-day campaign against universities: requests, tools, infrastructure Anything on the FBI breach, three months later
Trend Micro (18 June) How the exploit works inside PeopleSoft Any mention of the FBI
Google Threat Intelligence (25 Sept), as summarised by BleepingComputer and SC Media A renewed PeopleSoft campaign and a one-letter trick to dodge firewall rules Any mention of the FBI

Read together, they leave a hole in the middle. The FBI describes a platform and a missing patch without naming either. Oracle, Google and Trend Micro describe a PeopleSoft flaw, its fix and the group’s use of it, without mentioning the FBI. Only ShinyHunters joins the two, and it is the least reliable source in the table.

Who the FBI blames

The FBI said little for a fortnight, and each statement moved one step. It told Nextgov on 23 September that the cause was undetermined, and on X the next day that the breach was either at a third party or in its own enterprise (Fox News). An internal memo reported by the New York Times on 28 September was blunter: “We are operating under the premise that the threat actor is also exfiltrating” the personal data of all FBI employees (Gizmodo).

On 29 September Brett Leatherman, Assistant Director of the FBI’s Cyber Division, said in a video, covered by BleepingComputer, that Dutch police had arrested a 24-year-old in Amsterdam on 15 September, a week before the break-in, as an alleged leader, and that the group had breached more than 140 organisations and extorted at least $70 million. His message to the rest, as NPR reported: “You know how to find us, and we know how to find you.”

On 5 October the FBI told The Register it had “already worked with partners to arrest multiple subjects”. The admission the FBI had avoided came in a statement from Leatherman, reported by Reuters on 5 and 6 October: a “security failure of a platform managed by a third-party organization”, in which “a contractor failed to implement a security patch explicitly issued” (SecurityWeek). The FBI had “removed the contractor” on Monday 5 October (Security Affairs, citing Reuters). On 9 October Director Kash Patel announced the arrest of an accused co-conspirator, said the breach “occurred on a platform managed by a third-party vendor”, and promised to “disrupt what’s left of the ShinyHunters group and their associates” (AP).

So the victim makes the attribution, at director level. The public evidence behind it is short, and none of it is technical:

  • the group claimed the breach, with a defacement nobody disputes;
  • the leaked sample is genuine FBI data, as far as reporters could check;
  • the FBI says arrests followed, and links them to its investigation.

Arrests show the FBI knows who it is looking for. They don’t show how anyone got in. Google, which tracks the group as UNC6240, attributes the university campaign to it but hasn’t tied the group’s PeopleSoft work to the FBI.

The FBI hasn’t named the product, the vulnerability or the contractor. Accenture’s name comes from two sources familiar with the matter who spoke to Reuters, relayed by SecurityWeek, CSO and Nextgov; the same sources named PeopleSoft. Nobody reported that the FBI ended Accenture’s contract. Accenture said only that it was “proud to support the mission of the FBI and will continue to do so”.

What the group claimed, and what reporters checked

ShinyHunters said it had data on “all FBI employees and applicants”. What reporters have seen is far smaller, and still bad. 404 Media and Nextgov reviewed a sample covering about 5,000 employees: names, home addresses, phone numbers, and details of spouses and siblings. Nextgov found roles the FBI would never publish, including analysts working on Russia, China and Hezbollah, electronic surveillance staff, the Remote Operations Unit and the FISA Management Unit (FISA is the US law on foreign intelligence surveillance). Malwarebytes reported medical fitness-for-work records and blood tests in the leak, which the BBC said appear genuine.

Nobody has checked the rest: the total volume, the group’s figure of 60,000 staff, and access to systems the group called MedLink and BEAST. By 28 September the group was telling Nextgov it would not publish the data at all. The one claim that can be tested is the way in, because it names a product and a component.

How the attack worked

In the order the sources give, the breach has five steps. Two are confirmed by someone other than ShinyHunters (the way in, in outline, and the defacement), and one is partly verified. If Attack Flows are new to you, Beyond the ATT&CK Matrix: How to Build Dynamic Attack Flows with STIX explains how to read one.

  1. On the night of 21 September the group gets into the contractor-managed platform (Exploit Public-Facing Application, T1190). Confirmed in outline: the FBI says a contractor missed a patch. The group says a PeopleSoft zero-day; Reuters’ sources say PeopleSoft but name no flaw.
  2. It moves into FBI-managed AWS GovCloud. This is the group’s claim alone, and no source says how, so no technique fits. TTPs Are Missing the P: Let’s Fix That explains how to describe a step that no ATT&CK technique fits.
  3. It takes employee and applicant records (Data from Information Repositories, T1213). Partly verified through the sample.
  4. It defaces apply.fbijobs.gov (External Defacement, T1491.002). Confirmed.
  5. It gives the FBI a week to retract the May advisory. No technique describes a demand to withdraw a public warning.

The flow leaves out the group’s tooling, deliberately. On other PeopleSoft servers, according to BleepingComputer’s summary of Google’s September report, the group dropped JSP web shells (small server-side pages that run commands), a tunnelling page, a backdoor Google tracks as SIDEEYE, and MeshAgent, the MeshCentral remote-management agent. Nobody has said any of it was found at the FBI, so it appears below only as leads for hunters.

A patch three months old

Step 1 is where the two stories split, and the leading suspect is CVE-2026-35273. Oracle’s alert of 10 June describes a flaw in the Updates Environment Management component of PeopleTools 8.61 and 8.62, “remotely exploitable without authentication” and rated 9.8 out of 10 for severity. That component is the Environment Management Hub, PSEMHUB, which manages updates across PeopleSoft environments.

Trend Micro took the chain apart on 18 June. The hub’s validateClient check trusts requests that come from the server itself. The Integration Gateway (PSIGW), the HTTP listener PeopleSoft uses to swap messages with other systems, can be made to send a request to that local address on an attacker’s behalf, a server-side request forgery. Through it the attacker writes a file under envmetadata/, and when the service restarts, Java’s XMLDecoder turns the file back into objects and runs the attacker’s code. Trend Micro calls the result “behaviorally quiet”.

Google had already watched it used. Its 11 June report says ShinyHunters exploited it as a zero-day from 27 May to 9 June, in activity “consistent with” this CVE. Of more than 100 organisations Google notified, 68% were in higher education. The requests went to POST /PSEMHUB/hub and /PSIGW/HttpListeningConnector. The US Cybersecurity and Infrastructure Security Agency (CISA) added it to its Known Exploited Vulnerabilities (KEV) catalogue, the list of flaws US federal civilian agencies must patch by a set date, on 12 June with a 15 June deadline.

If this is the patch the FBI meant, the timeline is stark. By the group’s timing the break-in was the night of 21 September: 103 days after Oracle’s alert and 98 days after CISA’s deadline. That rests on an assumption the FBI hasn’t confirmed, but everything public fits it: a patch “explicitly issued”, a PeopleSoft platform according to Reuters’ sources, and a group that had exploited this exact flaw three months earlier. The flaw really was a zero-day, in May. By September it was a patch nobody had applied.

%50 is just a P

On 25 September Google reported that ShinyHunters was back at PeopleSoft with a twist, which BleepingComputer summarised: requests to /%50SEMHUB/ instead of /PSEMHUB/. %50 is URL encoding for the letter P. A web application firewall (WAF) rule written to block /PSEMHUB/ compares strings, sees no match and lets the request through. WebLogic decodes the path before routing it, sees /PSEMHUB/, and hands the request to the hub.

This is a parser differential: two components read the same request and disagree about what it says, the same root cause as HTTP request smuggling. MITRE files the bug class as CWE-180, validating input before canonicalising it, and the fix never changes: validate the request after decoding, as the application will see it. Google’s warning, as SC Media reported it, was that “organizations cannot assume placing a WAF in front of PSEMHUB is sufficient”: a WAF rule in place of Oracle’s fix is a control one encoded letter wide.

ShinyHunters told BleepingComputer it used this trick on FBIJobs.gov, along with a “NEW unknown vulnerability in the same PSEMHUB component”. The trick only helps against a server that is still vulnerable, which fits the FBI’s missed patch. A new unknown vulnerability would be a true zero-day, which doesn’t. And, as SC Media noted, the FBI was not mentioned in Google’s report.

What would settle it

The threat intelligence firm ThreatMon set out three explanations on 30 September: a new zero-day, a way around the CVE-2026-35273 protections, or a failure at the third party. The detection vendor Vectra AI had argued for the second: that the group got past WAF rules on servers that were never patched, and that “Organizations that had installed the patch were protected”. The FBI’s statement then pointed at the third without ruling out the second. One fact would settle each:

  • the FBI or the contractor naming the missed patch would confirm or rule out CVE-2026-35273;
  • Google tying its September activity to FBIJobs.gov would put the %50 trick at the FBI;
  • a new Oracle fix for PSEMHUB would show the “unknown vulnerability” exists.

The GovCloud move, and which systems beyond FBIJobs.gov were reached, wait on the FBI in the same way. Until it speaks, CVE-2026-35273 is the leading suspect for the way in, and nobody official has named it.

What defenders can hunt for

The FBI has published no indicators for this breach. The ones that exist come from Google’s research on the group’s university campaign in May and June, and nobody has tied any of them to the FBI, so treat them as leads on ShinyHunters rather than evidence of this breach: azurenetfiles[.]net (MeshCentral command and control), 176[.]120[.]22[.]24 (the mirror of the group’s leak site, which its staging server connected to) and the five staging servers 142[.]11[.]200[.]186 to 142[.]11[.]200[.]190. Google also gives SHA-256 hashes for four MeshCentral agents and the attackers’ shell history. All are in the reference section at the end. On PeopleSoft web servers, BleepingComputer’s summary of Google’s September report also gives file names worth searching for: x.jsp, u.jsp, u2.jsp, tunnel.jsp or tunnel.jspx, and Ple64.exe (not in the reference table).

IP addresses rotate, but every exploitation request Google and Trend Micro describe goes to the hub at /PSEMHUB/hub or the gateway at /PSIGW/HttpListeningConnector. A Sigma rule is a YAML file that describes suspicious activity in a log, which detection tools convert into queries for a specific SIEM. If you are new to Sigma, Writing Effective Sigma Detection Rules: A Guide for Novice Detection Engineers explains each field. This one is for the web server or reverse-proxy logs in front of PeopleSoft, and for WebLogic’s own access logs:

title: External or Loopback POST to PeopleSoft Environment Management Hub or Integration Gateway
id: 9b31656d-3738-5661-94a4-44f01a2c5e69
status: experimental
description: Detects HTTP POST requests from any address outside the private ranges, including loopback, to the PeopleSoft Environment Management Hub (PSEMHUB) or the Integration Gateway listener (PSIGW HttpListeningConnector). ShinyHunters exploited CVE-2026-35273 through POST /PSEMHUB/hub and /PSIGW/HttpListeningConnector in May and June 2026, and in September 2026 sent requests to /%50SEMHUB/hub, a percent-encoded spelling that path-matching WAF rules miss. Matching EMHUB/hub catches both spellings. The forged loopback request in Trend Micro's chain appears only in WebLogic's own access logs.
references:
    - https://cloud.google.com/blog/topics/threat-intelligence/shinyhunters-targets-education-sector-oracle-exploit
    - https://www.trendmicro.com/en/research/26/f/PeopleTools.html
    - https://www.bleepingcomputer.com/news/security/shinyhunters-uses-waf-bypass-trick-in-oracle-peoplesoft-attacks/
author: dogesec
date: 2026-10-09
tags:
    - attack.initial-access
    - attack.t1190
logsource:
    category: webserver
detection:
    selection_method:
        cs-method: 'POST'
    selection_path:
        cs-uri-stem|contains:
            - 'EMHUB/hub'
            - '/PSIGW/HttpListeningConnector'
    filter_internal:
        c-ip|cidr:
            - '10.0.0.0/8'
            - '172.16.0.0/12'
            - '192.168.0.0/16'
    condition: all of selection_* and not filter_internal
falsepositives:
    - Integration Broker (PeopleSoft's messaging layer) partners that post messages to HttpListeningConnector from fixed external addresses; add those addresses to the filter
    - Vulnerability scanners probing for CVE-2026-35273, which still means the endpoint is reachable from outside
level: high

The point is in the match. EMHUB/hub survives both the plain and the %50 spelling, and Sigma matches regardless of case, so the rule works whether your logs record the path as it arrived or after decoding. It is still a string match, like the WAF rules the group bypassed: an encoded E or H, or any encoded letter in /PSIGW/, slips past it, so match on the decoded path if your pipeline stores it. Don’t add the loopback range, because in Trend Micro’s chain the forged request to the hub comes from the server itself; that request never passes a reverse proxy, so only WebLogic’s own access logs will show it.

The rule covers step 1, and only if the way in was this flaw or a variant of it. Trend Micro’s chain finishes inside the Java process, and no web log shows a file being turned back into code. On the host, Google suggests looking for tar, zstd, rsync, sshpass and curl processes spawned from PeopleSoft or WebLogic service accounts, as SC Media reported, which would catch the group packing and moving data on other victims. Nothing covers the GovCloud move, the theft or the demand.

The whole case in one graph

Put together, this is the whole case: the reports, the FBI and who it blames, the five steps, the suspect flaw kept to one side, the group’s university campaign with its infrastructure, and the rule. If you are new to STIX, Understanding STIX 2.1 Objects explains what each node is.

A hunter gets two endpoints to watch and a short list of the group’s infrastructure, each marked as not tied to the FBI breach. An analyst gets the state of the case: the FBI blames ShinyHunters and a contractor’s missed patch, the product and the contractor are named only by Reuters’ sources, and CVE-2026-35273 fits every public fact without anyone official confirming it.

In summary

ShinyHunters’ defacement of the FBI’s jobs site produced 61 posts in 18 days, and most repeated either the group’s “zero-day” or, from 6 October, the FBI’s “missed patch”. The primary record is narrower: a victim admitting a contractor missed a patch, a PeopleSoft flaw fixed more than three months before the break-in, and research on a one-letter trick that slips past the firewall rules meant to stand in for that fix. The pieces fit, and nobody official has joined them. Defenders don’t need to wait: patch PeopleSoft rather than filtering it, and watch the two endpoints the group keeps coming back to.

Reference: IoCs and ATT&CK techniques

IoCs

All come from Google’s report on the May and June university campaign, and none is confirmed for the FBI breach.

Type Value What it is Linked to Confidence STIX ID
Domain azurenetfiles[.]net MeshCentral command and control, chosen to mimic Azure NetApp Files; not confirmed for the FBI breach ShinyHunters; MeshCentral 50 indicator--3ed4f7cb-793c-5726-add7-5a0d850835c0
IPv4 176[.]120[.]22[.]24 Clearnet mirror of the group’s leak site, reached over SSH from its staging server; not confirmed for the FBI breach ShinyHunters 50 indicator--d30a78ef-d36f-5b0f-ad22-c0c481642b2c
IPv4 142[.]11[.]200[.]186 Staging server (Python SimpleHTTP on port 8888); not confirmed for the FBI breach ShinyHunters 50 indicator--1431efcc-c090-53a3-8f23-16fca5f57619
IPv4 142[.]11[.]200[.]187 Staging server (Python SimpleHTTP on port 8888); not confirmed for the FBI breach ShinyHunters 50 indicator--f72563d5-b912-5f3d-8ac9-4497d5b58390
IPv4 142[.]11[.]200[.]188 Staging server (Python SimpleHTTP on port 8888); not confirmed for the FBI breach ShinyHunters 50 indicator--5c8b45cf-009e-5148-bb54-9438fb07bf9e
IPv4 142[.]11[.]200[.]189 Staging server (Python SimpleHTTP on port 8888); not confirmed for the FBI breach ShinyHunters 50 indicator--6631a5e5-4698-5d1e-a0f3-b60269478de9
IPv4 142[.]11[.]200[.]190 Staging server (Python SimpleHTTP on port 8888); not confirmed for the FBI breach ShinyHunters 50 indicator--9790ef48-cc7f-56fe-8262-7aa0f464cbdf
SHA-256 f02a924c9ff92a8780ce812511341182c6b509d45bc59f3f7b522e37225d24fc meshagent64-azure-ops.exe, preconfigured Windows MeshCentral agent; not confirmed for the FBI breach ShinyHunters; MeshCentral 50 indicator--dc1202d3-e80b-5c21-8530-b3287aaa08a0
SHA-256 d83fdb9e53c5ff03c4cb0451ea1bebd79b53f29eadc1e2fa394c7af13a86ce2f meshagent64-v2.exe, preconfigured Windows MeshCentral agent; not confirmed for the FBI breach ShinyHunters; MeshCentral 50 indicator--0ab05073-bb44-5bd0-b2cb-af5df05f4056
SHA-256 c7e9332731b06644fc73e0046a2a89eaa59b09f54250e9bd622467187351711f meshagent32-azure-ops.exe, preconfigured Windows MeshCentral agent; not confirmed for the FBI breach ShinyHunters; MeshCentral 50 indicator--deee593a-e68a-57ea-ab9d-487a40faffc9
SHA-256 68257a6f9ff196179ec03624e849927f26599eb180a7c82e14ef5bc4e93bc309 meshagent, unconfigured Linux MeshCentral agent; not confirmed for the FBI breach ShinyHunters; MeshCentral 50 indicator--e8f02582-2c07-5788-9963-07e30728a763
SHA-256 2ab684d93c1553fad87041b4dea97188a97e78589deee2a7bacff905564f3a35 .bash_history, attacker command history from the staging servers; not confirmed for the FBI breach ShinyHunters 50 indicator--650d2925-4f9f-543e-84ec-e8274fb9b88b

Techniques (FBIJobs.gov breach Attack Flow)

Step Action Technique Tactic STIX ID
1 Way in to the contractor-managed jobs platform T1190 TA0001 attack-action--2eb1ed64-733b-5669-a390-828681c5ded6
2 Claimed move into FBI-managed AWS GovCloud None (procedure) TA0008 procedure--bdec3696-2a69-5231-b751-c46ce9f48ecb
3 Employee and applicant records taken T1213 TA0009 attack-action--50f37d4f-9f7d-507c-b04c-9aeeedb0e5d3
4 apply.fbijobs.gov defaced T1491.002 TA0040 attack-action--953973cf-31f9-5069-95a2-467d739c56f3
5 Retraction demand with a one-week deadline None (procedure) TA0040 procedure--97ed1545-eb8f-56cd-8c3a-ffe618239bcb

The Sigma rule is indicator--9b31656d-3738-5661-94a4-44f01a2c5e69, the FBIJobs.gov campaign is campaign--8033cff8-06fa-5d2a-acd9-523bf989473d, and CVE-2026-35273 is vulnerability--c21a0d42-b85a-5a61-946e-ddfd655b3076.

Posted by:

Dylan Thomas

Dylan Thomas, Do Only Good Everyday

Get new posts by email

New threat research and tutorials from dogesec, plus the weekly Obstracts threat brief. No spam, unsubscribe any time.

Thanks. Check your inbox to confirm your subscription.