🔍 Read the full analysis: The AI Story In A $1,279 Exploit And A Lab Running Twice As Fast on ThorstenMeyerAI.com
Get smart everyday buys 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 a mid-October test, including 588 flaws it says were previously unknown. METR separately estimates that Anthropic’s model-development pace has accelerated, with AI likely contributing substantially. Key details are based on OpenAI’s reporting, while METR says much of its underlying evidence is not public.
OpenAI said its restricted GPT-6 Cyber model helped generate 948 working Linux privilege-escalation exploits during a mid-October test, while independent evaluator METR reported on November 5 that Anthropic’s AI-development pace has accelerated. The disclosures point to two related developments: AI tools are being used to find and exploit software flaws, and AI may be helping labs improve their own models faster. The figures come with significant limits: OpenAI supplied the exploit results, and METR says much of the evidence behind its estimate is not public.
OpenAI reported that its Linux project scanned about 2.8 million lines of kernel code, roughly 7% of the project. The company says its process identified 2,308 candidate findings, reproduced 1,724 vulnerabilities in test machines and produced 948 working local privilege-escalation exploits. OpenAI says 588 flaws were previously unknown. Those are company-reported results; the supplied material does not include an independent audit of the full run.
OpenAI named GPT-6 Cyber as the model behind the work on November 5. It is a cyber-specialized version of GPT-6 Astra, intended for authorized security testing, vulnerability research and web application testing. Access is limited to organisations approved for OpenAI’s Daybreak Red programme, which covers offensive security work. The company says access involves controls such as identity checks, hardware security keys and restrictions on API use.
The reported API bill was $1,213,430, or about $1,279 per working exploit using OpenAI’s totals. The source compares this with an earlier OpenAI run in July, which reported 166 Linux privilege-escalation bugs at a cost of $513,690, around $3,100 per exploit. That comparison concerns the two reported runs; it does not establish the cost or success rate for other systems or users.
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 cost matters because it suggests AI-assisted security testing could make large-scale vulnerability research more accessible to organisations that can afford the tools. It also raises the risk that capabilities used for authorized testing could be repurposed. A lower reported cost per exploit does not by itself show how reliably the method works across software, or whether attackers can reproduce the results.
For software maintainers and organisations, faster discovery can compress the time available to verify a flaw, develop a fix and distribute it before others try to exploit it. The supplied material says Mozilla issued public advisories crediting OpenAI with 50 high-impact Firefox fixes in October, but those advisories do not identify which model found each flaw. The practical implication for users is direct: apply security updates promptly, while organisations should monitor vendor advisories and maintain a process for patching exposed systems.
The METR report adds a separate concern about the pace of change. If AI tools help researchers build improved AI systems, development could accelerate in a feedback loop. METR estimates that AI accounts for much of Anthropic’s measured acceleration, but that is an evaluator’s estimate, not a fully public account of Anthropic’s internal work.
Products Worth Considering
Apple iPad mini (A17 Pro), 128GB, Wi-Fi 6E, Space Gray
Dell UltraSharp 27 Thunderbolt Hub Monitor (U2726DER), QHD
Dell UltraSharp 32 U3226QER 4K Thunderbolt Hub Monitor
As an affiliate, we earn on qualifying purchases.
Two Reports, Different Evidence
OpenAI’s announcement and METR’s assessment concern different companies and different kinds of evidence. OpenAI described a security exercise and supplied counts and costs. METR examined Anthropic’s model-development data over eleven weeks and concluded that the growth rate of its main measure was about 1.9 times faster after October 2025 than before. METR estimated that AI assistance likely explains about 80% of the acceleration it observed. That estimate should not be treated as a direct measurement of every part of Anthropic’s research process.
There is some outside information about security findings, but it does not verify every claim in OpenAI’s account. Mozilla’s public advisories credit OpenAI with high-impact Firefox fixes, while the source material says they do not connect particular fixes to a specific model. Epoch AI also reported that the freely downloadable model GLM 5.3 could turn published Firefox fixes into working attacks for about $1,000 per Firefox version. That is a separate reported demonstration, not independent validation of OpenAI’s Linux results.
OpenAI classifies GPT-6 Cyber as “Critical” under its own Preparedness Framework. That is the company’s rating, rather than a rating issued by an outside testing body. The source material also says no Linux kernel maintainer had commented publicly on OpenAI’s findings at the time of publication.
““A majority of the evidence we found most informative” was redacted.”
— METR, on the transparency of its Anthropic assessment
Independent Verification Remains Limited
The Linux totals and the $1,213,430 cost are OpenAI’s reported figures. The material provided does not show an independent reproduction of the 948 exploits, explain how each result was classified, or establish whether all 588 previously unknown flaws were disclosed and fixed. It also does not include a public response from Linux kernel maintainers.
For Firefox, Mozilla’s advisories provide evidence of credited fixes but do not identify which model discovered individual vulnerabilities. OpenAI’s access safeguards are described by the company; the available information does not show how well they have worked in practice or whether the reported model can be misused outside approved settings.
METR’s estimate is also qualified by its own disclosure: much of the evidence it considered most informative was withheld. The report offers an assessment of Anthropic’s pace and a likely explanation for the change, but readers cannot inspect most of the underlying data. It is not yet clear how much of the acceleration will persist or whether similar patterns apply across other AI labs.
Security Disclosure and Further Review
OpenAI’s Daybreak Red access is expected, according to reports cited in the source material, to reach approved Red customers over the coming weeks. The company says it can monitor, pause or revoke access, but further public information would be needed to assess the programme’s operation and safeguards.
For the Linux findings, the next useful developments would be public responses from maintainers, details on which flaws have been patched, and independent evaluation of the methods and results. For Firefox, advisories and technical disclosures may clarify the relationship between reported fixes and AI-assisted discovery.
On Anthropic, further evidence or updates from METR could clarify whether the estimated acceleration continues and how much AI contributes. In the meantime, the reports support a measured conclusion: AI-assisted vulnerability work is advancing, but the public record is not yet detailed enough to verify every headline figure or predict how quickly the changes will affect real-world security.
Key Questions
What did OpenAI report about the Linux test?
OpenAI said its process produced 948 working local privilege-escalation exploits after analysing Linux kernel code. It said 588 of the related flaws had not been known before. These figures are attributed to OpenAI and have not been independently verified in the material provided.
How was the $1,279-per-exploit figure calculated?
It is the reported $1,213,430 API bill divided by 948 working exploits. It is a calculation based on OpenAI’s totals, not a general estimate of the cost of finding or exploiting software flaws.
Who can use GPT-6 Cyber?
OpenAI says access is restricted to organisations it approves for the Daybreak Red programme. The programme is for offensive security testing, and the company describes identity, device and API restrictions for participants.
What did METR conclude about Anthropic?
METR estimated that Anthropic’s main model-development measure grew about 1.9 times faster after October 2025 than before. It estimated that AI likely accounts for around 80% of the acceleration it observed, while warning that much of its supporting evidence is not public.
What should organisations and users do now?
Organisations should track security advisories and apply tested patches quickly. Individual users should install operating-system and browser security updates promptly. The reports do not establish that any particular system is currently vulnerable.
Source: ThorstenMeyerAI.com
