🚨 ACTIVE EXPLOITATION
- Attackers exploit Rejetto HFS CVE-2026-61500 session forgery flaw for admin access and RCE
CVE-2026-61500
Attackers are actively exploiting CVE-2026-61500, a critical session-forgery flaw in Rejetto HFS discovered with Anthropic's Mythos model.- Affects Rejetto HFS 3.0.0 through 3.2.0 (the Node.js/TypeScript rewrite); fixed in version 3.2.1, released July 2026.
- The flaw (CVSS 9.3) lets attackers forge administrator session cookies because HFS derives its session signing key from JavaScript's Math.random() and leaks outputs of the same generator to unauthenticated clients during login.
- Attackers collect login responses, reconstruct the V8 xorshift128+ PRNG state (using the Z3 solver), recover the signing key, and achieve RCE via the server_code configuration feature.
- VulnCheck detected exploitation starting October 1, 2026, a day after Horizon3.ai's write-up; an actor hosted in China targeted vulnerable hosts in the US and Japan, with later hits from US proxy IPs 173.239.211[.]248 and 173.239.211[.]249.
- The bug was found by Horizon3's Zach Hanley using Anthropic's Mythos model via Project Glasswing; a public Python PoC by researcher aramosf appeared in late September 2026.
🕵️ RESEARCH & DEEP DIVES
Google's PageBreak AI agent found 500+ verified XSS flaws in its own web apps
Google's internal PageBreak AI agent has found and validated more than 500 cross-site scripting vulnerabilities across its first-party web applications.- Applies to Google's first-party web applications, including sensitive domains like apis.google.com and admin.google.com; this was defensive testing, not an intrusion.
- PageBreak, built by Google's Product Security team, began as a pilot in November 2025 and became a full project in January 2026, primarily using Gemini 3.1 Pro and Gemini 3.5 Flash models.
- It pairs LLM-driven bug hypotheses with deterministic validators that execute real payloads (e.g., injecting JavaScript and confirming execution in a rendering harness), yielding a near-zero false-positive rate; the same approach covers SQL injection, path traversal, RCE, and SSRF.
- Notable findings include a cache-poisoning XSS on apis.google.com via an unchecked URL path segment excluded from the cache key, a chained XSS on an admin console endpoint bypassing a cryptographic signature via a separate authorization flow, and a universal XSS in the Tag Assistant Extension via data URLs.
- Applications built on Google's high-assurance web frameworks showed only 2 XSS issues across hundreds of apps, both limited to internal apps or debug endpoints; no evidence of in-the-wild exploitation was found.
Sources 📄 Source: blog.google · 📎 Coverage: cybersecuritynews.com · 👁 via Cyber Security News, cryptika.com (discovered)Google suspends OSS VRP product vulnerability submissions amid AI-generated spam surge
Google has suspended product vulnerability submissions to its OSS VRP bug bounty program after a flood of invalid AI-generated reports.- Applies to security researchers reporting flaws in Google-maintained open-source projects including Golang, Angular, Bazel, Protocol Buffers, and Fuchsia.
- Product vulnerability submissions were halted effective October 1, 2026; supply chain reports, prior submissions, the Patch Rewards Program, and the Cloud VRP remain open.
- Google cited a significant rise in automated submissions, the vast majority invalid, overwhelming engineers and maintainers; an update is promised in Q1 2027.
- The OSS VRP launched in August 2022 with rewards of $100 to $31,337; curl and Intel have similarly curtailed bug bounty programs amid AI slop reports.
Open Build Service flaw CVE-2026-56004 allows command execution via Mercurial argument injection
Fenrisk found a second command execution vulnerability in openSUSE's Open Build Service, now fixed, caused by Mercurial argument injection in the obs_scm service.- Affects openSUSE Open Build Service (OBS), the packaging platform used by openSUSE, VLC, Dell, and Intel; the flaw is tracked as CVE-2026-56004 with CVSS 10 and is now patched.
- The obs_scm/tar_scm service's hg backend appends the user-controlled revision parameter to 'hg update' without a '--' separator, letting values starting with a dash be parsed as Mercurial options.
- Attackers who control a package's _service file can inject Mercurial's --config option to set a pre-update hook that runs arbitrary shell commands on the source service server, risking package signing secret theft and supply chain compromise.
- The payload bypasses Python argparse's dash-prefix rejection because values containing a space are treated as plain arguments; the Git backend is not exploitable due to this check.
- It follows the March 2025 CVE-2024-22033 wget argument injection flaw in the same attack surface.
Fortbridge details WordPress libheif RCE exploit chain via authenticated HEIC upload
Fortbridge researchers demonstrated a working remote code execution chain against WordPress servers running vulnerable libheif versions through ImageMagick.- Affects WordPress sites whose PHP/ImageMagick image backend links vulnerable libheif; requires an authenticated account with upload capability (Author or higher).
- Exploits libheif heap overflow GHSA-x8r2-mggj-j6wr in the ISO/IEC 23001-17 uncompressed image decoder's mixed-interleave YCbCr path, plus address-disclosure flaw GHSA-2jg2-4ch7-h545.
- Attack works by uploading crafted HEIC images via the Media Library API; WordPress-generated derivative JPEGs leak library and heap addresses, then a profile-specific 'unci' image triggers a heap overflow that hijacks a C++ virtual call.
- Validated on Ubuntu 26.04 (WordPress 7.1.1, PHP-FPM 8.5.4, ImageMagick 7.1.2.18, libheif 1.21.2) and Debian 13 (WordPress 7.0, PHP-FPM 8.4.24, ImageMagick 7.1.1.43, libheif 1.19.8); code executes as the PHP-FPM account (www-data).
- Success rates: 6 of 8 fresh PHP-FPM parent processes on Ubuntu, 22 of 24 on Debian; fixed in libheif 1.23.3.
🔓 CVEs & KEV
- CVE-2026-103510 — Perforce P4 (Helix Core) — CVSS 9.5 — Authentication bypass via blank auth token in P4SearchP4 Search prior to 2026...
📋 ADVISORIES
- 📄 Source for CISA adds actively exploited Zammad flaws CVE-2026-102489 and CVE-2026-102490 to KEV catalog — cisa.gov