🔍 Read the full analysis: What AI's $1,279 Exploit And Faster Lab Signal In GPT-6 Cyber And METR 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, 588 of them previously unknown, at about $1,279 per exploit. Separately, evaluator METR reports that Anthropic’s measured model progress accelerated after October 2025 and estimates AI-assisted research may explain much of the change. The figures are not equally open to independent verification, and the two developments point to growing pressure on software defenders.
OpenAI named GPT-6 Cyber on November 5 as the model behind an October campaign that, according to the company, generated 948 working Linux privilege-escalation exploits, including 588 previously unknown flaws. On the same day, independent evaluator METR reported faster progress at Anthropic, saying its main measure of model capability grew about 1.9 times faster after October 2025 than before. Together, the reports suggest AI is improving both software vulnerability discovery and the research process used to build models, though key evidence remains restricted or company-reported.
OpenAI describes GPT-6 Cyber as a cyber-specialized version of GPT-6 Astra for authorized security work. Access is limited to organizations approved for the company’s Daybreak Red programme, which covers offensive security testing. OpenAI’s system card lists vulnerability discovery, exploit development and web application testing among the model’s intended uses. The company assigns it a “Critical” cybersecurity rating, the highest tier in its own Preparedness Framework; that rating is OpenAI’s assessment, not an independent certification.
OpenAI says the Linux campaign scanned about 2.8 million lines of kernel code, identified 2,308 candidate findings and reproduced 1,724 vulnerabilities in test machines. The company reports 948 working local privilege-escalation exploits after grouping findings by root cause, with 588 described as previously unknown. Its stated API bill was $1,213,430, or roughly $1,279 per working exploit. The run occurred in mid-October; OpenAI identified GPT-6 Cyber as the model behind it on November 5.
METR’s separate report says Anthropic’s main measure of model progress grew 1.9 times faster after October 2025 than in the preceding period. METR estimates that AI assistance to Anthropic researchers may account for about 80% of the acceleration. That is an estimate, not a directly observed breakdown: METR says Anthropic withheld much of the evidence it found most informative. The report therefore offers an external assessment, but not a fully open audit of the underlying data.
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
Shorter Timelines for Security Teams
The reported exploit output matters because it suggests vulnerability research can be conducted at a scale and cost that would have been difficult for many teams to match manually. The comparison offered in the source material is with an earlier OpenAI Linux run in July: 166 privilege-escalation bugs for $513,690, or about $3,100 per exploit. OpenAI’s October figures imply a lower stated cost per exploit, although both sets of figures are company-reported and may not represent equivalent work or independently verified results.
For software maintainers and organizations, faster discovery can mean a shorter window between finding a flaw and needing to fix it. The same capabilities can support legitimate audits and defensive testing, but exploit-building tools could also create risks if access controls fail or similar capabilities become widely available. The practical challenge is not only finding flaws; it is reviewing reports, validating them, issuing patches and getting those patches installed promptly.
METR’s findings add a second pressure point. If AI tools are helping researchers improve models faster, capability changes may arrive more quickly than organizations’ security policies and evaluation processes can adapt. METR’s estimate does not prove that AI caused the acceleration, but it makes the pace of AI-assisted research a matter for ongoing scrutiny.
Products Worth Considering
Bose SoundLink Flex Bluetooth Speaker (2nd Gen) – Portable Outdoor Speaker with Hi‑Fi Audio, Waterproof and Dustproof, USB‑C, Up to 12 Hours
LG UltraGear 32GS60QC-B 32-inch Curved QHD Gaming Monitor (180Hz, 1ms, 1000R, AMD FreeSync, HDR10)
LG UltraGear 34GX90SA-W 34-inch WQHD Curved OLED Gaming Monitor
As an affiliate, we earn on qualifying purchases.
How the Two Reports Connect
The reports concern different organizations and different evidence. OpenAI’s announcement describes the output of a specific security exercise and its own model. METR evaluates Anthropic’s progress over time. They should not be treated as a joint study or as proof that one development caused the other. Their connection is the broader possibility that AI is advancing both the ability to identify software weaknesses and the speed of AI development itself.
OpenAI’s access restrictions are part of its stated response to the dual-use nature of GPT-6 Cyber. Daybreak Red organizations must meet requirements including single sign-on, multifactor authentication and managed devices; individual researchers face identity checks and hardware-key requirements, according to the system card. OpenAI says it monitors usage and can suspend or revoke access. These measures describe the company’s controls, not an independent assessment of how well they work.
There is some outside documentation related to the security work. Mozilla’s public advisories credit OpenAI with 50 high-impact Firefox fixes in October, according to the supplied source material. Those advisories do not identify which model found each issue. Separately, Epoch AI reported that a freely downloadable model called GLM 5.3 could turn published Firefox fixes into working attacks for about $1,000 per Firefox version. That finding concerns a different model and task; it does not independently verify OpenAI’s Linux results.
““a majority of the evidence we found most informative””
— METR
Evidence Still Behind Access Limits
The Linux exploit count, the number of previously unknown flaws and the API cost come from OpenAI’s account of its own campaign. The supplied material does not establish that an independent party reproduced all 948 exploits or verified the full cost calculation. It also says no Linux maintainer had commented publicly at the time of the report. The extent to which the findings have been disclosed, patched or accepted by maintainers is not established here.
Mozilla’s advisories provide a separate, public record of fixes credited to OpenAI, but do not connect individual flaws to GPT-6 Cyber. METR’s assessment is independent, yet its report has a different transparency limitation: Anthropic redacted much of the evidence METR considered most informative. The estimate that AI explains about 80% of Anthropic’s acceleration should be read as METR’s judgment, not as a measured causal share.
The material also does not establish whether the reported reduction in cost per exploit holds across other software, models or security tasks. Nor does it show how frequently the restricted model’s safeguards block misuse. Those questions will depend on further disclosures, independent testing and the public record of patches and advisories.
Disclosure, Patches and Independent Checks
The next evidence to watch is the handling of the Linux findings: whether maintainers or security advisories confirm affected components, what fixes become available, and how many systems need updates. OpenAI says GPT-6 Cyber is available only to approved Daybreak Red organizations, with access expected to extend to eligible customers over the coming weeks, according to reporting cited in the source material. The company’s stated restrictions and monitoring will remain part of the scrutiny around the release.
Further public documentation could clarify how the Linux findings were validated and how the campaign’s cost was calculated. METR’s future evaluations may also show whether Anthropic’s faster measured pace continues and whether more of the evidence can be made public. Until then, the reports support a cautious conclusion: AI-assisted security work is advancing, but the size, reliability and broader effects of the claims are not yet fully independently established.
For organizations and individual users, the immediate response remains practical rather than speculative: track vendor security advisories, test and apply updates promptly, and maintain processes for triaging vulnerability reports. The reports do not show that every system is exposed, but they underline the importance of reducing the time between a verified flaw and a deployed fix.
Key Questions
What is GPT-6 Cyber?
OpenAI describes it as a cyber-specialized version of GPT-6 Astra intended for authorized security work, including vulnerability discovery and exploit testing. Access is restricted to organizations approved for its Daybreak Red programme.
Did OpenAI independently verify the 948 Linux exploits?
The figure comes from OpenAI’s account of its October campaign. The supplied material does not establish that an independent party reproduced all 948 exploits or confirmed the full set of results.
What did METR find about Anthropic?
METR reported that its main measure of Anthropic’s model progress grew about 1.9 times faster after October 2025 than before. It estimated AI assistance may account for about 80% of the acceleration, while acknowledging that much of the evidence was redacted.
Does the METR report verify OpenAI’s security claims?
No. METR’s report concerns the pace of progress at Anthropic. It is separate from OpenAI’s claims about GPT-6 Cyber and the Linux exploit campaign.
What should organizations and users do now?
Follow security advisories from software vendors, prioritize the testing and installation of relevant patches, and keep vulnerability-response processes ready to handle reports quickly. The supplied material does not identify a specific Linux patch or a confirmed exposure for every system.
Source: ThorstenMeyerAI.com
