93← Home

OpenAI's new GPT-6 Cyber AI finds 81 Firefox flaws, 948 Linux bugs

Headline from BleepingComputer · compiled from 6 sources

Our analysis · checked against 12 sources

OpenAI's restricted GPT-6 Cyber model built 948 Linux exploits, found 81 Firefox flaws

OpenAI says its GPT-6 Cyber model, released November 5 to approved customers, produced 948 working local Linux root exploits from 2,308 candidates, versus 166 for GPT-5.6 Sol, and 81 new Firefox flaws Mozilla rated high severity. Fixes lag: of 304 Linux findings disclosed, 61 were fixed upstream within 10 days; how many of the 948 were previously unknown is disputed.

The framing"Finds 81 Firefox flaws, 948 Linux bugs" does match OpenAI's counts, and Mozilla rated all 81 Firefox flaws high severity. But the 948 are working local exploits verified by OpenAI's own pipeline; only 304 Linux findings have been disclosed to maintainers, and 61 were fixed upstream within 10 days.

What the headline leaves out

  • The 948 are local privilege-escalation exploits: attackers need existing access, not remote entry.
  • Only 304 Linux findings were disclosed; 61 fixed upstream and 43 staged within 10 days.
  • A rival model, Claude Mythos, was reported to find 271 Firefox zero-days earlier this year.
  • The run cost $1,213,430 in API credits; the model is paid and access-restricted.

What happens next

OpenAI says it plans to expand access to all Daybreak Red customers 'soon', with no date given. Remediation is incomplete: OpenAI's Fouad Matin points to fewer than 25% of flaws fixed on its Patch the Planet dashboard, and further Linux disclosures are underway. Dragos's Rob T. Lee warns 'the skill floor is dropping' for attack research. Watch the dashboard's fixed share and any independent re-verification of the findings.

Possible knock-on effects
  • Lower skill needed for vulnerability research could let less-skilled attackers find exploitable flaws, which may raise pressure on defenders and operators of shared or untrusted-workload systems. (named expert: Rob T. Lee, Dragos Chief AI Officer: 'The skill floor is dropping'. Watch for: Reports of attacks using AI-found local privilege-escalation flaws)
  • With 948 exploits but only 304 findings disclosed, further disclosures could add to maintainer workload, and unpatched flaws may stay exposed on shared Linux systems until fixes reach releases. (mechanism: Follows from the gap between exploits built and findings disclosed and fixed; no source makes this forecast. Watch for: Patch the Planet dashboard fixed share rising above 25%, or Linux maintainers commenting on disclosure volume)

Sources: OpenAI: Introducing GPT-6 Cyber (primary), OpenAI GPT-6 Cyber system card (primary), Cyber Security News: OpenAI launches GPT-6 Cyber, How-To Geek: OpenAI cyber model gained Linux root access, Tech Times: GPT-6 Cyber finds 81 Firefox vulnerabilities, 948 Linux exploits, Benzinga: GPT-6 Cyber uncovered 588 unknown Linux vulnerabilities, Unite.AI: OpenAI launches GPT-6 Cyber with restricted access, BleepingComputer: GPT-6 Cyber finds 81 Firefox flaws, 948 Linux bugs, The Next Web: GPT-6 Cyber launch, Daybreak Red, Startup Fortune: 2,308 Linux kernel bugs into 948 root exploits, OrcaRouter: GPT-6 Cyber vs MAI Cyber, MarketScale: GPT-6 Cyber for approved security teams

Written in our own words from the sources above, and checked claim by claim before it was published. The report below says only what each outlet said.

The report: what outlets said

BleepingComputer reports that OpenAI has revealed GPT-6 Cyber, a new AI model that has discovered 81 previously unknown high-severity Firefox vulnerabilities and developed working exploits for 948 distinct Linux kernel flaws.[1]

OpenAI describes GPT-6 Cyber as its "cybersecurity model for finding exploitable vulnerabilities and developing fixes, trained to stay within authorized scope."[2]According to Cyber Security News, OpenAI's Linux pipeline produced 2,308 candidate findings and reproduced 1,724 of them. It then built 948 working local privilege escalation exploits, which let a normal user become root.[3] Benzinga reports that OpenAI disclosed 304 critical findings to maintainers along with proposed patches.[4]On Firefox, Brian Grinstead, a Distinguished Engineer at Firefox, told Help Net Security that the model "uncovered previously unknown high-severity vulnerabilities, identified architectural improvements, and helped us ship fixes to millions of users"[5]. Cyber Security News reports that Mozilla's advisory MFSA 2026-118 covers fixes in Firefox 159, released October 27.[3]According to Help Net Security, GPT-6 Cyber scored 48.5% on ExploitGym, while its predecessor, GPT-5.6 Cyber, scored 32.1%.[5] The Hacker News reports that OpenAI has classified the model as Critical in cybersecurity under its Preparedness Framework.[6]BleepingComputer reports that the new model is not publicly available, with access currently restricted to a select group of approved customers in OpenAI's Daybreak Red.[1]OpenAI also told BleepingComputer: "It is clear that we have a limited window to use frontier models to secure software."[1]

Listen to the songAI News — November 07, 2026

Sources

  1. [1]OpenAI's new GPT-6 Cyber AI finds 81 Firefox flaws, 948 Linux bugs
  2. [2]GPT‑6 Cyber: Advancing the frontier of aligned cybersecurity agents | OpenAI
  3. [3]OpenAI Launches GPT-6 Cyber That Finds 81 Firefox Flaws and Develops 948 Linux Exploits
  4. [4]OpenAI's New GPT-6 Cyber AI Can Find and Exploit Software Flaws — It Already Uncovered 588 Unknown Linux - Benzinga
  5. [5]OpenAI’s GPT-6 Cyber found 81 Firefox flaws earlier AI testing missed - Help Net Security
  6. [6]OpenAI Launches GPT-6 Cyber After AI Helps Find 81 Firefox Flaws, 948 Linux Exploits

Every sentence here reports what a named outlet said, with a link to it. Compiled by 93 from the reporting below.

This topic, day by day

OpenAI's new GPT-6 Cyber AI finds 81 Firefox flaws, 948 Linux bugs | 93