🔍 Read the full analysis: The AI Signals Behind A $1,279 Exploit And A Lab Moving Twice As Fast on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
OpenAI says GPT-6 Cyber helped produce 948 working Linux privilege-escalation exploits during an October test, including 588 flaws it says were previously unknown. Separately, METR estimates Anthropic’s measured AI progress has accelerated to about 1.9 times its earlier pace, while warning that much of the supporting evidence is confidential.
OpenAI said its GPT-6 Cyber model helped produce 948 working Linux privilege-escalation exploits in a test run in mid-October, including 588 flaws the company describes as previously unknown. On November 5, the company also announced restricted access to the cyber-focused model, as independent evaluator METR reported that Anthropic’s measured AI progress has accelerated to roughly 1.9 times its earlier pace. The claims point to pressure on software defenders and to AI’s growing role in the research process, but much of the evidence behind both developments is not public.
OpenAI’s November 4 account of the Linux campaign says the work covered about 2.8 million lines of kernel code, roughly 7% of the project. Its pipeline identified 2,308 candidate findings, reproduced 1,724 vulnerabilities in test machines and generated 948 working local privilege-escalation exploits, after grouping findings by root cause. OpenAI says 588 of the 948 were previously unknown. Those figures come from the company; the source material does not include an independent audit of the complete results or public comment from Linux kernel maintainers.
OpenAI named the model behind the run as GPT-6 Cyber, a cyber-specialized version of GPT-6 Astra intended for authorized security work. Access is limited to organizations approved for the Red tier of its Daybreak programme, which covers offensive testing such as penetration tests and red teaming. OpenAI’s system card sets out requirements including multifactor authentication, managed devices and restricted API access. The company rates the model “Critical” for cybersecurity, the top tier in its own Preparedness Framework; that is OpenAI’s assessment, not an outside certification.
METR’s separate report says its main measure of Anthropic’s model capabilities improved about 1.9 times faster after October 2025 than before. METR estimates that AI may account for about 80% of the acceleration by helping Anthropic researchers, but emphasizes that this is an estimate. The evaluator says Anthropic redacted a majority of the evidence it found most informative, limiting what readers can independently verify.
GPT-6 Cyber · the Linux run
948 working exploits, about $1,279 each
OpenAI’s agents worked through the Linux kernel in mid-October. Every number on this board comes from OpenAI; our own arithmetic is marked ≈.
API cost per working Linux privilege-escalation exploit in OpenAI’s October run: $1,213,430 for 948 exploits. OpenAI’s own figures.
- Candidate findings analysed2,308
- Reproduced in test machines1,724
- Working exploits948
- …previously unknown588
About 2.8 million lines of kernel code scanned (≈7%). 948 = unique flaws after grouping 1,169 exploits by root cause.
Three months, one curve
OpenAI Linux runs, cost per privilege-escalation bug
- July 2026 · GPT-5.6 Sol166 bugs for $513,690≈ $3,094
- October 2026 · GPT-6 Cyber948 exploits for $1,213,430≈ $1,279
- Past hacking-competition payoutsper Linux privilege-escalation exploit$20k–70k
What is and is not known
As of 6 Nov 2026
- Reports go to kernel maintainers with candidate patches; a human checks each one first (OpenAI).
- Technical details, including at least 10 new techniques, are withheld until fixes ship.
- No public statement from Linux kernel maintainers found yet.
- Validated flaws with suggested patches are not the same as shipped fixes.
Sources: OpenAI, “A thousand fewer paths to root”, 4 Nov 2026, as quoted by The Hacker News and Decrypt; 588 previously unknown and the model as reported by Fortune, CSO Online and The Register, 5 Nov 2026. ≈ values: own arithmetic.
Powered by Thorsten Meyer AI
Cheaper Exploit Discovery Pressures Defenders
The reported Linux results matter because they suggest that finding and turning software flaws into working exploits may require less time and money than it did in earlier AI-assisted tests. The source material compares an OpenAI July run that produced 166 Linux privilege-escalation bugs for $513,690—about $3,100 per exploit—with the October results, which work out to about $1,279 each using OpenAI’s figures. These are company-reported results from different runs, not proof that every vulnerability can be found or exploited at that price.
Lower costs could benefit security teams testing their own systems, but the same capabilities could increase pressure on maintainers and organizations to identify and patch weaknesses quickly. The source also cites Epoch AI’s late-October report that a freely downloadable model, GLM 5.3, used published Firefox fixes to produce working attacks for about $1,000 per Firefox version. That finding suggests access to advanced tools is not limited to restricted commercial services, though the details and conditions of that work matter.
METR’s estimate adds a second concern for defenders and the public: if AI tools help researchers improve AI systems faster, capability changes may arrive more quickly than external evaluations and safeguards can track. For users, the practical implication remains straightforward: install security updates promptly. The larger organizational challenge is matching a faster discovery cycle with timely triage, fixes and deployment.
Popular right now
Samsung 65-Inch The Frame Pro Neo QLED 4K TV (2026) Bundle with Modern Brown Bezel, HDMI Cables, and Home Theater Guide
De'Longhi Dinamica Plus Fully Automatic Espresso Machine
Samsung The Frame Pro LS03HW 65-inch Neo QLED TV with 27-Month Axiom Care
As an affiliate, we earn on qualifying purchases.
From Kernel Tests to Restricted Access
OpenAI’s Linux disclosure followed an earlier test in July, which it says found 166 privilege-escalation bugs at an API cost of $513,690. The October run produced a higher count at a lower reported cost per exploit, but the source material does not establish that the tests used identical methods, coverage or definitions. The figures should be read as results from two company-reported experiments rather than a controlled measure of performance over time.
OpenAI also reported that its work found serious Firefox vulnerabilities. Mozilla’s public advisories credit OpenAI with 50 high-impact Firefox fixes in October, while OpenAI’s account refers to 81 previously unknown high-severity flaws. The advisories do not identify which model discovered each issue, so they do not independently confirm GPT-6 Cyber’s contribution to specific bugs. OpenAI says the Linux findings were shared for remediation; the supplied material does not detail the full disclosure timeline or the status of every flaw.
GPT-6 Cyber is OpenAI’s fourth cyber-focused model release this year, according to the source material. Its Daybreak programme has Blue access for defensive work and Red access for offensive testing; this model is restricted to organizations OpenAI approves for Red. Separately, METR’s report examines the pace of progress at Anthropic using its own measure, not OpenAI’s cybersecurity benchmarks. The two developments are related by the broader role of AI in research, but they are separate findings based on different methods and evidence.
““The public has to rely on METR’s judgment much more than would be ideal.””
— METR, on the limits of its Anthropic assessment
Key Results Still Lack Public Verification
The Linux totals and the claim that 588 flaws were previously unknown are OpenAI-reported figures. The supplied material does not provide a public, independent review of the complete set of findings, explain how every exploit was validated, or say whether all issues have been fixed. No Linux maintainer had commented publicly at the time described in the source.
Mozilla’s advisories document fixes credited to OpenAI, but they do not show which model found which vulnerability. It is also unclear how the 81 Firefox flaws cited by OpenAI relate to the 50 high-impact fixes in Mozilla’s public advisories. The information provided does not reconcile those counts or establish that they refer to the same categories or reporting period.
METR’s 1.9-times figure is an evaluator’s result, while its estimate that AI contributed about 80% of the acceleration is an interpretation of the available evidence. Because Anthropic redacted much of that evidence, outsiders cannot fully reproduce the analysis from the public report. The two reports do not establish that AI progress will continue at the same rate or that the same pace applies across labs.
Disclosure, Patches and Access Decisions
OpenAI says GPT-6 Cyber is available through its API and Codex to approved Daybreak Red organizations, with access expected to expand to additional eligible customers over the coming weeks, according to the reporting cited in the source material. The company says it monitors use and can pause or revoke access. The next evidence to watch is how the model performs in authorized security work and whether OpenAI publishes more detail on evaluation and safeguards.
For the Linux and Firefox findings, the key next steps are remediation, public advisories and clearer accounting of which vulnerabilities were verified and fixed. Maintainer statements or independent technical analysis could help validate the scale of the Linux results. Organizations should continue applying vendor security updates quickly while avoiding assumptions that every flaw in the reported totals remains exploitable or unpatched.
METR’s findings may be updated as further evidence becomes available. Any additional disclosure from Anthropic or METR could clarify how the evaluator calculated its pace comparison and the estimated AI contribution. For now, the confirmed public development is a pair of reports with different evidence limits: OpenAI’s company-reported security results and METR’s assessment, which itself flags substantial redactions.
Key Questions
What did OpenAI say GPT-6 Cyber found?
OpenAI says the October Linux test produced 948 working privilege-escalation exploits, including 588 flaws it describes as previously unknown. The figures have not been independently verified in the source material.
Can the public use GPT-6 Cyber?
No. OpenAI limits access to organizations it approves for the Red tier of its Daybreak programme, for authorized offensive security testing. The company says access is provided through its API and Codex.
What does METR say about Anthropic’s progress?
METR reports that its main measure of Anthropic’s model progress improved about 1.9 times faster after October 2025 than before. It estimates AI may account for about 80% of the acceleration, but much of the underlying evidence is confidential.
Are the Linux and Firefox findings independently confirmed?
Not in full, based on the supplied material. Mozilla advisories credit OpenAI with Firefox fixes, but do not identify which model found each flaw. The Linux totals are OpenAI’s figures, and no public comment from Linux maintainers was reported.
What should computer users do now?
Install operating-system, browser and software security updates promptly. The reports do not show that every reported flaw is currently exploitable or unpatched, but faster vulnerability discovery makes timely updates important.
Source: ThorstenMeyerAI.com
