🔍 Read the full analysis: What A $1,279 Exploit And Faster Lab Mean For AI's Direction on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
OpenAI says GPT-6 Cyber helped produce 948 working Linux privilege-escalation exploits in an October run, including 588 flaws it had not previously known. Separately, METR estimates Anthropic’s model-development pace increased about 1.9 times after October 2025, with AI likely contributing substantially. The figures point to faster security discovery and AI research, but much of the supporting evidence remains private.
OpenAI says its newly named GPT-6 Cyber helped generate 948 working Linux privilege-escalation exploits in a mid-October test, including 588 flaws the company said were previously unknown. On November 5, independent evaluator METR also reported that Anthropic’s model-development pace had increased about 1.9 times after October 2025, with AI itself likely responsible for much of the acceleration.
OpenAI described the Linux campaign in a November 4 post and identified GPT-6 Cyber as the model behind it the following day. The company says its pipeline examined about 2.8 million lines of Linux kernel code, assessed 2,308 candidate findings and reproduced 1,724 vulnerabilities in test machines. It reported 948 working local privilege-escalation exploits after grouping results by root cause. The figures are OpenAI’s own account; independent confirmation of the Linux findings is not included in the supplied material.
The campaign’s reported API bill was $1,213,430, or about $1,279 per working exploit. OpenAI cited 588 previously unknown flaws among the 948. That is a raw cost-per-exploit calculation, not a measure of the cost to validate, disclose or fix each vulnerability. The account also gives no independent audit of the billing or the exploit count.
GPT-6 Cyber is a cybersecurity-focused version of GPT-6 Astra, according to OpenAI, and is restricted to organisations it approves for its Daybreak Red programme. The system card says it is intended for authorised security work, including vulnerability research and penetration testing. Separately, METR said Anthropic’s measured progress accelerated after October 2025; it estimated that roughly 80% of the acceleration may come from AI assistance to Anthropic researchers, while warning that much of its supporting evidence is not public.
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 Exploits Tighten Patch Timelines
If OpenAI’s figures hold up, the cost of producing working exploits has fallen sharply in its own tests: the source reports a July campaign that found 166 Linux privilege-escalation bugs for $513,690, compared with 948 exploits for about $1.21 million in October. Those figures imply a lower reported cost per exploit, but the campaigns’ comparability and the results have not been independently established here.
The practical concern is not only whether an AI system can identify flaws, but how quickly defenders can verify and repair them before others use the same techniques. OpenAI says GPT-6 Cyber access is restricted, yet a separate late-October Epoch AI report cited in the source says a freely downloadable model, GLM 5.3, produced working attacks from published Firefox fixes at a reported cost of about $1,000 per Firefox version. That is a separate claim and does not establish that the model found new vulnerabilities.
For organisations, the developments raise the pressure to monitor vulnerability reports, prioritize patches and test software quickly. For individual users, the immediate advice remains straightforward: install security updates promptly. The claims do not show that ordinary users are facing a specific new attack, or that every discovered flaw is exploitable outside a test environment.
Products Worth Considering
Apple MacBook Air 13-inch Laptop with M5 Chip, 16GB Memory, 512GB SSD, 12MP Center Stage Camera, Touch ID, Wi-Fi 7, Midnight
Apple MacBook Air 15-inch with M5 chip, 16GB memory, 512GB SSD, and Liquid Retina display (Midnight)
Logitech MX Keys Combo for Business Keyboard and Mouse – Graphite
As an affiliate, we earn on qualifying purchases.
Restricted Model, Public Security Findings
OpenAI describes GPT-6 Cyber as its fourth cyber-focused model released in 2026, following versions announced in April, June and August, according to the source article. The model is available through the API and Codex only to organisations approved for Daybreak Red, the programme tier intended for offensive security testing. OpenAI says access requires organisational controls such as multi-factor authentication and managed devices, and that it can monitor, suspend or revoke use.
OpenAI rates GPT-6 Cyber “Critical” for cybersecurity under its own Preparedness Framework. That is the company’s assessment, not an external certification. Its developer pricing page, as described in the source, lists $20 per million input tokens and $100 per million output tokens for approved organisations.
Mozilla’s public advisories reportedly credit OpenAI with 50 high-impact Firefox fixes in October. They do not identify which model found each issue. The source also describes an earlier OpenAI Linux run in July that reported 166 privilege-escalation bugs at a cost of $513,690. Separately, METR’s analysis focuses on Anthropic’s internal research pace, not on the security performance of GPT-6 Cyber.
“A majority of the evidence we found most informative”
— METR, describing evidence limits in its Anthropic report
Evidence Behind the Numbers Remains Limited
The Linux exploit totals, number of previously unknown flaws and API costs are attributed to OpenAI. The supplied material does not provide an independent audit, a list of the Linux findings or public confirmation from Linux maintainers. It is also unclear how many of the reported flaws have been fixed, how broadly they affect deployed systems, or whether the campaign’s results can be reproduced by independent researchers.
Mozilla’s advisories confirm that OpenAI received credit for 50 high-impact Firefox fixes, according to the source, but do not establish which model discovered them. The source says no Linux maintainer had commented publicly at the time of publication. METR’s estimate that AI accounts for about 80% of Anthropic’s acceleration is also an estimate, and METR said Anthropic withheld much of the evidence it found most informative.
The source article notes that it was drafted with help from Anthropic’s Claude, while Anthropic and OpenAI are subjects of the report and competitors. That disclosure does not settle the accuracy of the claims; readers should distinguish company figures, evaluator estimates and independently published advisories.
Access Rollout and Patch Verification
OpenAI’s Daybreak Red access is expected to reach approved customers over the coming weeks, according to reports cited in the source. The company says access is limited to vetted organisations and can be paused or revoked. Further details on the Linux findings, including maintainer responses and public patch status, would help establish the practical impact of the campaign.
Security teams and users should watch for vendor and project advisories and apply relevant updates as they become available. The source does not identify a specific release schedule for Linux fixes or a date for additional independent checks of OpenAI’s totals. METR’s estimate will also remain difficult to assess from outside while much of its underlying evidence is withheld.
Key Questions
What did OpenAI say GPT-6 Cyber found?
OpenAI reported 948 working Linux local privilege-escalation exploits from a mid-October campaign and said 588 involved previously unknown flaws. These totals have not been independently verified in the supplied material.
How much did the reported Linux campaign cost?
OpenAI’s reported API bill was $1,213,430, which works out to about $1,279 per working exploit. That figure does not include other costs such as independent validation, disclosure or remediation.
Can anyone use GPT-6 Cyber?
No. OpenAI says it is restricted to organisations it approves for its Daybreak Red programme, which is intended for authorised offensive security work.
What did METR say about Anthropic?
METR estimated that Anthropic’s measured model-development pace was about 1.9 times faster after October 2025 than before. It estimated that AI may account for roughly 80% of the acceleration, while noting that much of its most informative evidence is not public.
What should computer users do now?
Install security updates promptly and follow advisories from the software and operating-system providers you use. The reports do not identify a specific new attack against ordinary users.
Source: ThorstenMeyerAI.com
