# OpenAI GPT-6 Cyber Identifies 81 Previously Unknown High-Severity Firefox Vulnerabilities (Mozilla MFSA2026-118, Firefox 159)

> GBHackers reports that OpenAI's GPT-6 Cyber model produced evidence and reproducible tests for 81 novel Firefox flaws that Mozilla classified as high severity, and found 30 known issues that were worse than first reported. Mozilla shipped fixes in Firefox 159 (MFSA2026-118); no in-the-wild exploitation is reported.

- **Published:** 2026-11-06T05:06:19Z
- **Last reviewed:** 2026-11-06T05:06:19Z
- **Canonical:** https://intel.threadlinqs.com/threat/TL-2026-5373
- **Card image:** https://intel.threadlinqs.com/og/threat/TL-2026-5373.png
- **ID:** TL-2026-5373
- **Severity:** HIGH
- **Category:** VULNERABILITY
- **Status:** MONITORING
- **Detections:** 9 (1 shown free below) · **IOCs:** 16
- **CVEs:** CVE-2026-117876, CVE-2026-117878, CVE-2026-117879, CVE-2026-118008, CVE-2026-117881, CVE-2026-117882, CVE-2026-117883, CVE-2026-117921, CVE-2026-117922, CVE-2026-117930, CVE-2026-117931, CVE-2026-117932, CVE-2026-117933, CVE-2026-117934, CVE-2026-118011, CVE-2026-118012, CVE-2026-117935, CVE-2026-117936, CVE-2026-117937, CVE-2026-117938, CVE-2026-117939, CVE-2026-117940, CVE-2026-117941, CVE-2026-117942, CVE-2026-117943, CVE-2026-117944, CVE-2026-117945, CVE-2026-117946, CVE-2026-117947, CVE-2026-117948, CVE-2026-117949, CVE-2026-117950, CVE-2026-117951, CVE-2026-117953

## Description

GBHackers (2026-11-06) reports that OpenAI's GPT-6 Cyber, a cybersecurity-specialised model distributed to vetted defenders through OpenAI's Daybreak programme, found 81 previously unknown Firefox vulnerabilities that Mozilla rated high severity. Each finding came with evidence and reproducible tests. The model also found 30 already-known Firefox issues that proved worse than first reported. Mozilla began landing fixes soon after receiving the findings and proposed patches. The article reports no in-the-wild exploitation.

The same article cites Mozilla advisory MFSA2026-118, which Mozilla published on 2026-10-27 for Firefox 159. We fetched that advisory. It covers 126 individually numbered CVEs: 94 high, 29 moderate and 3 low impact. Most high-impact entries are use-after-free conditions in the DOM, Layout, JavaScript engine (GC), Graphics (Canvas2D), WebRTC, Audio/Video, Storage, XML and XSLT components. The advisory also lists a same-origin-policy bypass (CVE-2026-117860), a sandbox escape (CVE-2026-117861) and a privilege escalation from incorrect boundary conditions in Networking (CVE-2026-117862). Mozilla states that it changed its methodology to issue an advisory for every individual bug instead of consolidating memory-safety bugs into one CVE.

Attribution caveat: the advisory does not name GPT-6 Cyber. It credits high-impact CVEs to the reporters 'OpenAI Codex Security (kylebot)' (components include CSS parsing, WebSockets, DOM bindings/events/core, Layout, XML, SVG, JS engine and JIT) or Amy Burnett of OpenAI (Layout, IndexedDB, DOM Navigation). These are the closest sourced mapping to OpenAI-originated work, but they are not confirmed to be the 81 in the GBHackers claim, and the advisory gives no 81-finding count. Data-quality caveat: repeated fetches of the advisory returned inconsistent summary totals (126, 161 and 166 CVEs) and slightly different credited-CVE lists, so cve_list (34 IDs) is the union of the OpenAI-credited IDs seen across fetches and must be reconciled against the live advisory before detection content is derived from it. The OpenAI announcement (openai.com/index/introducing-gpt-6-cyber/) returned HTTP 403 and was not read.

Broader context from the GBHackers article: GPT-6 Cyber reportedly produced 948 Linux kernel exploits, reproduced 1,724 Linux kernel vulnerabilities and disclosed 304 critical findings to maintainers. It scored 48.5% on ExploitGym versus 32.1% for GPT-5.6 Cyber, and reportedly avoided honeypots 99.93% of the time. Access sits behind the Daybreak Red and Daybreak Blue tiers, with identity verification, SSO/MFA, FIDO2 keys and IP allowlisting. Other outlets report that OpenAI rates the model at a 'critical' cyber capability level under its Preparedness Framework. Earlier 2026 reporting describes comparable AI-assisted Firefox work by Anthropic (Claude Opus 4.6: 22 vulnerabilities, 14 high severity, fixed mostly in Firefox 148; Claude Mythos Preview: 271 vulnerabilities fixed in Firefox 150).

Defensive relevance: the main risk is faster, cheaper discovery and exploit development for memory-safety bugs in browsers and kernels, which shortens patch windows. Defenders should keep Firefox current (159 or later) and watch for post-patch n-day exploitation. No IOCs, exploit code or attacker infrastructure are published.

## MITRE ATT&CK

- T1588.006 Vulnerabilities
- T1587.004 Exploits
- T1203 Exploitation for Client Execution

## Free detection (Splunk SPL, Microsoft KQL, Sigma)

**TL-2026-5373 Firefox Spawning Shell or Script Interpreter (MFSA2026-118 Exploitation)** (`TL-2026-5373-SPL-001`)

Detects the Firefox main or content process (firefox.exe, including -contentproc children) spawning cmd, PowerShell, WSH or WSL. Firefox never needs to start these interpreters in normal browsing, so a hit is a strong post-exploitation signal of a memory-corruption exploit (e.g. the use-after-free class fixed in Firefox 159, MFSA2026-118, CWE-416) reaching code execution. No in-the-wild exploitation is reported yet, so this is an n-day readiness detection for unpatched Firefox (< 159).

- **Languages:** Splunk SPL, Microsoft KQL, Sigma
- **ATT&CK:** T1203, T1189, T1059
- **Data sources:** sysmon, microsoft_defender_for_endpoint
- **Severity:** high
- **Confidence:** high

### Splunk SPL

```spl
index=sysmon earliest=-24h sourcetype="XmlWinEventLog:Microsoft-Windows-Sysmon/Operational" EventCode=1 ParentImage="*\\firefox.exe" (Image="*\\cmd.exe" OR Image="*\\powershell.exe" OR Image="*\\pwsh.exe" OR Image="*\\wscript.exe" OR Image="*\\cscript.exe" OR Image="*\\bash.exe" OR Image="*\\wsl.exe")
| eval content_process=if(like(ParentCommandLine, "%-contentproc%"), "yes", "no")
| stats count min(_time) as first_seen max(_time) as last_seen values(CommandLine) as child_commandlines values(content_process) as spawned_by_content_process by host User ParentImage Image
| convert ctime(first_seen) ctime(last_seen)
| sort - count
```

### Microsoft KQL

```kql
DeviceProcessEvents
| where Timestamp > ago(24h)
| where InitiatingProcessFileName =~ "firefox.exe"
| where FileName in~ ("cmd.exe", "powershell.exe", "pwsh.exe", "wscript.exe", "cscript.exe", "bash.exe", "wsl.exe")
| extend ContentProcess = InitiatingProcessCommandLine has "-contentproc"
| project Timestamp, DeviceName, AccountName, ContentProcess, InitiatingProcessCommandLine, FileName, ProcessCommandLine, SHA256
| order by Timestamp desc
```

### Sigma

```yaml
title: Firefox Spawning Shell or Script Interpreter (TL-2026-5373)
id: 1b2c3d4e-5f60-4a71-8b92-a3b4c5d6e701
status: experimental
description: Detects firefox.exe (main or content process) spawning a command shell or script host, a strong indicator of browser exploitation such as the use-after-free flaws fixed in Firefox 159 (MFSA2026-118).
references:
  - https://www.mozilla.org/en-US/security/advisories/mfsa2026-118/
  - https://gbhackers.com/openais-new-gpt-6-cyber-finds-81-high-severity-firefox-vulnerabilities/
author: AII-Detector — ThreadLinqs Intelligence
date: 2026-11-06
tags:
  - attack.initial_access
  - attack.t1189
  - attack.execution
  - attack.t1203
  - attack.t1059
logsource:
  category: process_creation
  product: windows
detection:
  parent:
    ParentImage|endswith: '\firefox.exe'
  child:
    Image|endswith:
      - '\cmd.exe'
      - '\powershell.exe'
      - '\pwsh.exe'
      - '\wscript.exe'
      - '\cscript.exe'
      - '\bash.exe'
      - '\wsl.exe'
  condition: parent and child
falsepositives:
  - Browser extensions or native-messaging hosts that launch helper scripts
  - Enterprise kiosk or SSO tooling that launches scripts from Firefox
level: high
```

**Known false positives:**

- Firefox native-messaging hosts or extensions that intentionally launch helper scripts
- Enterprise kiosk, SSO or password-manager integrations that start PowerShell or cmd from Firefox

**Tuning notes:** Baseline for one week and allowlist known native-messaging host command lines via CommandLine exclusions. Raise priority when spawned_by_content_process=yes or when the host runs Firefox older than 159. Pair with the exposure inventory for CVEs from MFSA2026-118.

### 8 more detection rules (Splunk SPL, Microsoft KQL, Sigma), unlock with TL Creds

- TL-2026-5373 Repeated [indicator] Crashes in xul.dll (Memory-Corruption Exploit Attempts) (locked: unlock with TL Creds)
- TL-2026-5373 Discovery Commands Spawned from [indicator] (Post-Exploitation Recon) (locked: unlock with TL Creds)
- TL-2026-5373 Firefox-Descended Process Making Outbound Public Connections (C2 Callback) (locked: unlock with TL Creds)
- TL-2026-5373 Linux [indicator] Spawning Shell or Network Tool (Browser Exploitation) (locked: unlock with TL Creds)
- TL-2026-5373 [indicator] Writing Run Key or Startup Folder Persistence (locked: unlock with TL Creds)
- TL-2026-5373 [indicator] Spawning Living-off-the-Land Proxy Execution Binaries (locked: unlock with TL Creds)
- TL-2026-5373 [indicator] Sandbox Escape or Elevated Child Process (CVE-2026-117861 / CVE-2026-117862) (locked: unlock with TL Creds)
- TL-2026-5373 Executable or Script Dropped by [indicator] and Executed (Post-Exploit Ingress) (locked: unlock with TL Creds)

Sign up free to unlock the detection rules: https://threadlinqs.com/?modal=signup

## Sources

- [OpenAI's New GPT-6 Cyber Finds 81 High-Severity Firefox Vulnerabilities and 948 Linux Kernel Exploits](https://gbhackers.com/openais-new-gpt-6-cyber-finds-81-high-severity-firefox-vulnerabilities/)
- [Mozilla Foundation Security Advisory 2026-118 (Firefox 159)](https://www.mozilla.org/en-US/security/advisories/mfsa2026-118/)
- [Introducing GPT-6 Cyber (OpenAI; cited by GBHackers, HTTP 403 on fetch)](https://openai.com/index/introducing-gpt-6-cyber/)
- [OpenAI prepara GPT-6 Cyber para el DevDay del 29 de septiembre](https://wwwhatsnew.com/2026/09/28/openai-gpt-6-cyber-modelo-ciberseguridad-devday-2026)
- [OpenAI lleva GPT-6 Cyber a DevDay: defensa con IA](https://ecosistemastartup.com/openai-lleva-gpt-6-cyber-a-devday-defensa-con-ia/)
- [OpenAI GPT-5.6 Cyber announcement and Daybreak access (Gigazine)](https://gigazine.net/news/20260812-gpt-5-6-cyber/)
- [OpenAI is reportedly preparing to preview a GPT-6 Cyber security model](https://thenextweb.com/news/openai-gpt-6-cyber-preview-fortune)
- [Anthropic says it found a heap of Firefox security flaws using new Claude tools (context)](https://www.techradar.com/pro/security/anthropic-says-it-found-a-heap-of-firefox-security-flaws-using-new-claude-tools-says-ai-is-making-it-possible-to-detect-severe-security-vulnerabilities-at-highly-accelerated-speeds)

## Full data

The other 8 detection rules (Splunk SPL / Microsoft KQL / Sigma) unlock with TL Creds; a free account includes daily creds: https://threadlinqs.com/?modal=signup
Indicator values unlock with TL Creds, at a lower price on Red and Purple; AI agents can query rules and IOCs through the Threadlinqs MCP server (Purple): https://intel.threadlinqs.com/mcp

Canonical: https://intel.threadlinqs.com/threat/TL-2026-5373