Threat reportVulnerabilityTL-2026-2865
Rejetto HTTP File Server (HFS) 3.x session forgery via predictable Math.random() signing key leads to unauthenticated admin access and RCE (CVE-2026-61500) under active exploitation
Rejetto HFS 3.0.0 through 3.2.0 derives its session-cookie signing key from the non-cryptographic Math.random() generator and leaks outputs of the same generator to unauthenticated clients, letting a remote attacker recover the PRNG state, forge an administrator session cookie and execute code via the server_code feature.
Key facts
| Identifiers | TL-2026-2865, CVE-2026-61500, CVE-2026-61501 |
|---|---|
| Also known as | Rejetto HFS session forgery via predictable signing key |
| Severity | CRITICAL |
| CVSS | 9.8 (CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H) |
| EPSS | CVE-2026-61500: 0.9% (61.2th percentile); CVE-2026-61501: 0.4% (34.2th percentile) |
| Affected products | Rejetto HTTP File Server (HFS) |
| Weaknesses (CWE) | CWE-338, CWE-79, CWE-330, CWE-347, CWE-94 |
| Nation-state nexus | China |
| Tools | Z3 SMT solver |
| MITRE ATT&CK techniques | T1059.001 Command and Scripting Interpreter, T1059.003 Command and Scripting Interpreter, T1059.007 JavaScript, T1078 Valid Accounts, T1087 Account Discovery, T1090 Proxy, T1105 Ingress Tool Transfer, T1190 Exploit Public-Facing Application, T1212 Exploitation for Credential Access, T1550.004 Web Session Cookie, T1595.002 Active Scanning, T1606.001 Web Cookies |
| Motivation | UNKNOWN |
| Target regions | united states of america, japan |
| Detection rules | 9 (1 shown free) |
| Sources | 9 cited sources |
Detection coverage
As of 2026-10-12, Threadlinqs Intelligence publishes 9 detection rule(s) for TL-2026-2865 across Splunk SPL, Microsoft KQL and Sigma, covering 26 indicator(s) of compromise. The whole corpus is readable without an account. Every threat comes with one complete detection in Splunk SPL, Microsoft KQL and Sigma, free for everyone; a free account adds daily TL Creds to unlock the rest. Paid plans add a larger daily allowance, correlation and the MCP server. Threadlinqs MCP server · View plans.
Free detection: Rejetto HFS Admin API Used Without a Prior Successful Login (Forged Session Cookie)
Flags external sources that receive successful (HTTP 200) responses from HFS administrative API endpoints but have no successful login recorded from the same address in the window. A cookie forged with the recovered keygrip signing key is accepted without any login, which is exactly this pattern; the forged cookie lets the attacker operate as a valid administrator account (T1078) by replaying web session material (T1550.004). Localhost is excluded because HFS may grant admin access there by design.
- Languages: Splunk SPL, Microsoft KQL, Sigma
- ATT&CK: T1606.001, T1550.004, T1078
- Data sources: HFS access log (index=web sourcetype=hfs:access), HFSAccess_CL (custom Log Analytics table), webserver logs
- Severity: critical
- Confidence: medium
Splunk SPL
```CVE-2026-61500 forged admin session: admin API 200s with zero successful logins from same src```
index=web sourcetype="hfs:access" earliest=-24h src_ip!="127.0.0.1" src_ip!="::1"
| eval is_admin_api=if(match(uri_path, "^/~/api/(set_config|get_config|get_status|get_vfs|set_vfs|get_accounts|set_account|add_account|get_plugins|start_plugin)$") AND status=200, 1, 0)
| eval is_login_ok=if(match(uri_path, "^/~/api/(login|loginSrp2)$") AND status=200, 1, 0)
| stats sum(is_admin_api) AS admin_calls sum(is_login_ok) AS successful_logins min(_time) AS first_seen max(_time) AS last_seen values(eval(if(is_admin_api=1, uri_path, null()))) AS admin_endpoints by src_ip, host
| where admin_calls>0 AND successful_logins=0
| convert ctime(first_seen) ctime(last_seen)Microsoft KQL
let LoginOk = HFSAccess_CL
| where TimeGenerated > ago(24h)
| where UriPath_s in~ ("/~/api/login", "/~/api/loginSrp2") and Status_d == 200
| summarize LoginOkCount=count() by SrcIp_s, Host_s;
HFSAccess_CL
| where TimeGenerated > ago(24h)
| where UriPath_s matches regex "^/~/api/(set_config|get_config|get_status|get_vfs|set_vfs|get_accounts|set_account|add_account|get_plugins|start_plugin)$"
| where Status_d == 200
| where SrcIp_s !in ("127.0.0.1", "::1")
| summarize AdminCalls=count(), Endpoints=make_set(UriPath_s), FirstSeen=min(TimeGenerated), LastSeen=max(TimeGenerated) by SrcIp_s, Host_s
| join kind=leftanti LoginOk on SrcIp_s, Host_s
| project FirstSeen, LastSeen, SrcIp_s, Host_s, AdminCalls, EndpointsSigma
title: Rejetto HFS Administrative API Success From External Client (Possible Forged Session)
id: 2e7a9c04-81b6-4d3f-b5a2-7f60c3d14e95
status: experimental
description: Successful response from a Rejetto HFS administrative API endpoint to a non-internal client. Sigma cannot express absence of a prior login; pair with the SPL/KQL variants, which require zero successful logins from the same source, to isolate forged admin cookies (CVE-2026-61500).
references:
- https://horizon3.ai/attack-research/disclosures/anthropic-mythos-rejetto-hfs-rce/
author: Threadlinqs Team
date: 2026/10/03
tags:
- attack.credential-access
- attack.t1606.001
- attack.defense-evasion
- attack.t1550.004
- attack.initial-access
- attack.t1078
- cve.2026.61500
logsource:
category: webserver
detection:
selection:
cs-uri-stem:
- '/~/api/set_config'
- '/~/api/get_config'
- '/~/api/get_accounts'
- '/~/api/set_account'
- '/~/api/add_account'
- '/~/api/set_vfs'
- '/~/api/start_plugin'
sc-status: 200
filter_internal:
c-ip|cidr:
- '127.0.0.0/8'
- '10.0.0.0/8'
- '172.16.0.0/12'
- '192.168.0.0/16'
condition: selection and not filter_internal
falsepositives:
- Administrators managing an internet-exposed HFS instance from a remote address
level: highKnown false positives:
- Administrator whose browser holds a long-lived valid session cookie so no login appears in the 24h window
- Legitimate remote administration of an exposed HFS instance
Tuning notes: Extend the lookback beyond 24h (or join against a login lookup) to cut false positives from long-lived admin sessions. Allowlist known admin source IPs. Verify the successful-login endpoint names (login / loginSrp2) in your HFS version; if HFS logs the cookie or username, key the correlation on that instead of source IP.
- Rejetto HFS Probe Burst Followed by Admin API Access From Same Source (CVE-2026-61500 Chain) · unlock with TL Creds
- Rejetto HFS [indicator] Configuration Write From External Source · unlock with TL Creds
- Rejetto HFS Runtime Spawning Shell or LOLBin Child Process (Windows) · unlock with TL Creds
- Rejetto HFS API Access From Reported CVE-2026-61500 Attacker Subnet ([indicator]) · unlock with TL Creds
- Rejetto HFS config.yaml Rewritten by HFS Runtime ([indicator] Persistence) · unlock with TL Creds
- Rejetto HFS Node Process Spawning Shell or Downloader (Linux) · unlock with TL Creds
- Rejetto HFS Math.random() Output Harvesting - Unauthenticated Login Burst (CVE-2026-61500) · unlock with TL Creds
- Rejetto HFS Process Initiating Outbound Connection to Public IP (Post-Exploitation Egress/Proxy/Tool Transfer) · unlock with TL Creds
Indicators of compromise
As of 2026-10-12, Threadlinqs Intelligence tracks 26 indicators of compromise for TL-2026-2865.
Sample indicators (3 of 26)
- IP address:
173.239.211[.]248 - URL:
hxxps://github[.]com/rejetto/hfs/releases/tag/v3.2.1 - IP address:
173.239.211[.]249
Defanged, TLP:CLEAR. Full set of 26 indicators: unlock with TL Creds.
Technical analysis
VulnCheck canaries observed exploitation attempts from a China-based IP against real vulnerable hosts in the US and Japan, followed by activity from two US-based IPs in the same /24.
CVE-2026-61500 affects the TypeScript rewrite of Rejetto HTTP File Server (HFS), versions 3.0.0 through 3.2.0. HFS generates the key used to sign session cookies (via the Koa web framework's keygrip) with JavaScript's Math.random(). V8 implements Math.random() with xorshift128+, a non-cryptographic generator whose internal state is fully reversible. HFS additionally discloses raw Math.random() outputs to unauthenticated clients during login, so an attacker has the observations needed to reconstruct the generator state.
Per Horizon3.ai's write-up and The Register's summary, the exploit chain is: (1) user enumeration to confirm that an admin account exists; (2) collect roughly 12 consecutive Math.random() outputs from the unauthenticated endpoint that leaks them; (3) use the Z3 SMT solver to recover the xorshift128+ internal state; (4) step the state backwards to reconstruct the signing key generated at server startup; (5) forge a valid administrator session cookie (session IP binding could also be bypassed through forged cookie parameters); (6) abuse built-in administrative functionality, namely the server_code configuration feature that accepts custom JavaScript, to run arbitrary code on the host. No prior authentication is required against an internet-reachable instance.
The flaw was discovered by Zach Hanley of Horizon3.ai working with Anthropic's Mythos model (Project Glasswing). It is tracked as CWE-338. The CVE record was published by VulnCheck on 2026-07-13, coinciding with HFS 3.2.1, whose release notes state that multiple security vulnerabilities in all previous versions could allow an attacker to gain administrative access. Horizon3's technical disclosure followed in late September 2026. CVSS is 9.8 (CVSS 3.1, AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H) per OpenCVE, and 9.3 (CVSS 4.0) as assigned by the CNA, VulnCheck. Sources report different scores for the same flaw.
According to The Register, VulnCheck's Patrick Garrity said VulnCheck began detecting exploitation of CVE-2026-61500 on the evening of the story, and that canaries saw an actor in China targeting real vulnerable hosts in the US. Targeting of hosts in the US and Japan was reported. Two US-based IPs, 173.239.211.248 and 173.239.211.249 in the same subnet, were reported as follow-on attackers and are suspected to be proxies. The reporting calls the activity China-nexus without naming a group, and notes it is consistent with the documented use of compromised devices as routing proxies in an April 2026 multi-country advisory. The CVE is listed in VulnCheck's KEV database. CraftedSignal's brief records it as not on CISA KEV. The Register cites Patrick Garrity's tracker as showing it is only the second Mythos/Project Glasswing-credited CVE exploited in the wild, out of 286 tracked.
BeaconBeagle returned no record (HTTP 404) for either reported IP at time of research, so no C2 correlation is available.
MITRE ATT&CK techniques by tactic:
Execution
T1059.001 Command and Scripting Interpreter; T1059.003 Command and Scripting Interpreter; T1059.007 JavaScript
Privilege Escalation
Discovery
Command and Control
T1090 Proxy; T1105 Ingress Tool Transfer
Initial Access
T1190 Exploit Public-Facing Application
Credential Access
T1212 Exploitation for Credential Access; T1606.001 Web Cookies
lateral-movement
Reconnaissance
Remediation
Patches
- Rejetto HFS 3.2.1 (https://github.com/rejetto/hfs/releases/tag/v3.2.1)
Immediate actions
- Upgrade all Rejetto HFS instances to 3.2.1 or later
- Inventory HFS deployments and identify which are reachable from the internet; prioritise those for patching
- On any exposed pre-3.2.1 instance, review the server_code configuration and administrator accounts for unexpected changes
- Review web and authentication logs for repeated unauthenticated login requests and user enumeration from a single source
Workarounds
- Remove internet exposure of vulnerable instances until patched
Longer-term hardening
- Do not expose HFS administration to the internet; restrict access by network ACL or VPN
- Use cryptographically secure randomness (for example Node.js crypto.randomBytes()) for any security-sensitive value, never Math.random()
Affected and fixed versions
- Rejetto — HTTP File Server (HFS)
Vulnerable versions: 3.0.0 through 3.2.0
Fixed in: 3.2.1
Timeline
Last updated · 6 updates · revalidated 6× · latest source
- CVE-2026-61500 reserved following discovery by Zach Hanley of Horizon3.ai with Claude and Anthropic Research.
- HFS 3.2.1 also fixes CVE-2026-61501, a stored XSS in the admin log viewer via a crafted failed-login username (CVSS 5.3); VulnCheck publishes a separate advisory.
- Rejetto releases HFS 3.2.1, noting multiple security vulnerabilities in all previous versions that could allow administrative access; VulnCheck publishes the CVE-2026-61500 advisory the same day.
- Zach Hanley of Horizon3.ai, working with Anthropic's Mythos model under Project Glasswing, identifies the Math.random()/xorshift128+ session-key weakness (Horizon3 dates discovery to September 2026; exact day not given).
- Public Python PoC (aramosf/CVE-2026-61500) released in late September 2026 (exact date not given); it demonstrates state recovery, cookie forgery and server_code RCE against loopback.
- Horizon3.ai publishes its technical disclosure of the HFS session forgery to RCE chain (publication date as given on the Horizon3 page; The Register describes the publication as 'Wednesday').
- Active exploitation attempts detected from October 1 (VulnCheck, Patrick Garrity); NVD/CVE record updated the same day.
- VulnCheck records four further attempts from two US-based IPs in the same subnet (locked-ioc-4, locked-ioc-5), assessed as likely proxy infrastructure.
- VulnCheck warns of small-scale reconnaissance and exploitation attempts from a China Telecom IP against canaries in Japan and the US, and adds the CVE to its KEV catalog.
- The Register reports CVE-2026-61500 as the second Mythos/Project Glasswing-credited CVE (of 286 tracked) exploited in the wild.
- The Register reports further exploitation from two US-based IPs in the same subnet (locked-ioc-4 and locked-ioc-5), suspected to be proxies.
- VulnCheck reports its canaries detected a China-based IP exploiting CVE-2026-61500 against real vulnerable hosts in the US; targeting of hosts in the US and Japan is reported.
- CISA KEV catalog snapshot (1,734 entries) does not contain CVE-2026-61500; the CVE is listed only in VulnCheck's KEV database.
- BleepingComputer reports, citing VulnCheck Canary Intelligence honeypots, small-scale scanning from a single China Telecom IP against HFS deployments in Japan and the US, and notes no successful exploitation or post-exploitation activity was reported in that article.
- The Hacker News reports exploitation by an unnamed China-based actor against vulnerable U.S. hosts.
- eSecurityPlanet reports roughly 100 internet-facing HFS instances and active exploitation attempts; CISA KEV listing was not confirmed at that time.
- The Register reports Mythos's second in-the-wild-exploited vulnerability, exploited within about a day of PoC publication with narrow US/Japan targeting.
Update history
- 2026-10-12 — Rejetto HFS CVE-2026-61500 Session Forgery Auth Bypass to RCE (Predictable Math.random() Signing Key) Found with Anthropic Mythos and Probed in the Wild: What changed No field changes. Severity CRITICAL, exploitability ACTIVE and status ACTIVE are already at their maximum. The newer report's CVSS 4.0 9.3 is already recorded.
- 2026-10-10 — CVE-2026-61500: Rejetto HFS 3.x Session Forgery via Predictable Signing Key Leads to Admin Takeover and RCE (found by Horizon3 with Anthropic Mythos Preview): What changed No severity, exploitability, status or CVSS change. Existing values (CRITICAL, ACTIVE, 9.8 under CVSS 3.1) already match or exceed the new report. Its 9.3 is the CVSS 4.0 score, which the record already notes.
- 2026-10-05 — Rejetto HFS CVE-2026-61500: Math.random() Session-Key Prediction Enables Admin Cookie Forgery and RCE, Now Actively Scanned: What changed No field escalation. Severity (CRITICAL), exploitability (ACTIVE) and status (ACTIVE) are already at the top of their ranges. New CVEs (0) None. CVE-2026-61500 is already recorded.
- 2026-10-05 — Rejetto HFS CVE-2026-61500 session forgery to authentication bypass and RCE, exploited in the wild (found with Anthropic Mythos): What changed No severity, exploitability or status change; already CRITICAL/ACTIVE. Scope widened to include the companion XSS CVE-2026-61501 fixed in the same release. New indicators (2) Indicator values are withheld at this access level.
- 2026-10-05 — Active Exploitation of Rejetto HFS Authentication Bypass to RCE (CVE-2026-61500): What changed No change to severity (CRITICAL), exploitability (ACTIVE) or status. Report corroborates existing exploitation findings. New indicators (2) Indicator values are withheld at this access level.
- 2026-10-05 — Rejetto HFS CVE-2026-61500: Weak PRNG Enables Admin Session Forgery and RCE, Actively Exploited: What changed No field escalation: exploitability (ACTIVE), severity (CRITICAL) and attribution (China, LOW) already match. Added a public PoC and earlier detected-exploitation date (Oct 1).
Sources cited
- VulnCheck advisory: Rejetto HFS session forgery via predictable signing key vulncheck.com · Advisory · 2026-07-13
- NVD: CVE-2026-61500 NVD (NIST) · Advisory · 2026-07-13
- Horizon3.ai disclosure: Anthropic Mythos Rejetto HFS RCE horizon3.ai · Research · 2026-09-30
- Rejetto HFS v3.2.1 release notes GitHub · Vendor report · 2026-07-13
- OpenCVE: CVE-2026-61500 app.opencve.io · Research · 2026-07-13
- Halo Security CVE advisory: Rejetto HFS predictable session key allows RCE cve.halosecurity.com · Research · 2026-07-13
- CraftedSignal brief: Rejetto HFS session key feed.craftedsignal.io · Research · 2026-07-13
- Patrick Garrity: Anthropic-Credited CVEs tracker GitHub · Code repository
- The Register: Anthropic's super bug-hunting model Mythos is hardcore good at math, as latest vuln under attack shows The Register · News · 2026-10-03
Previous and next in vulnerability
Further reading
Drafted by the Threadlinqs research pipeline from 9 sources; published ; last updated .