🔍 Read the full analysis: A $1,279 Exploit And A Lab Running Twice As Fast: What GPT-6 Cyber And The METR Report Show 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 an AI campaign in October produced 948 working Linux privilege-escalation exploits, including 588 it says were previously unknown, at an API cost of about $1,279 per exploit. On November 5, OpenAI identified the model as GPT-6 Cyber, while METR reported that Anthropic’s measured AI capabilities are improving about 1.9 times faster than before. Both findings rely partly on information that outside readers cannot independently inspect.
OpenAI said GPT-6 Cyber powered an October campaign that produced 948 working Linux privilege-escalation exploits, including 588 the company says were previously unknown, for an API bill of $1,213,430. On November 5, independent evaluator METR also reported that Anthropic’s measured rate of AI capability improvement had accelerated, linking the two developments to growing AI contributions to both vulnerability research and model development.
OpenAI’s reported Linux pipeline began with scans of about 2.8 million lines of kernel code, or roughly 7% of the project. The company says its process examined 2,308 candidate findings, reproduced 1,724 vulnerabilities in test machines and generated 948 working local privilege-escalation exploits after grouping results by root cause. OpenAI attributed 588 of the 948 to previously unknown flaws. Those figures come from the company’s account; the source material does not include an independent audit of the complete campaign.
The API cost reported for the run was $1,213,430, equivalent to about $1,279 per working exploit. That is a cost calculation for this reported run, not a measure of the full cost of validating, reporting and fixing vulnerabilities. The source compares it with an earlier OpenAI run in July, which it says found 166 Linux privilege-escalation bugs for $513,690, or about $3,100 each. The figures suggest a lower API cost per exploit across the two runs, though the available account does not establish that the runs used identical methods or scope.
OpenAI named the model as GPT-6 Cyber on November 5. It describes the system as a cyber-specialized version of GPT-6 Astra intended for authorized security work, including vulnerability discovery, exploit development and web-application testing. Access is restricted to approved organisations in the company’s Daybreak Red programme, which covers offensive testing. OpenAI says GPT-6 Cyber is rated “Critical,” the highest level in its own cybersecurity preparedness framework; that designation is the company’s assessment, not an outside certification.
METR’s report examined Anthropic data over an 11-week period. It found that its main measure of AI capability growth rose about 1.9 times faster after October 2025 than before. METR estimated that AI assistance to Anthropic’s own researchers may account for roughly 80% of the acceleration. That share is the evaluator’s estimate, not a directly observed or independently reproduced measurement of every contribution.
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, Faster Patch Pressure
If the reported results generalize, AI is lowering the cost of finding and turning software flaws into working exploits. That can help authorised security teams discover weaknesses earlier, but it also means defenders may have less time between a vulnerability becoming findable and someone using it. The per-exploit API cost is not the total cost of a security incident or of remediation, but it points to a lower technical barrier for producing exploit candidates.
The practical effect depends on verification and response. A model-generated exploit still needs to be checked, disclosed responsibly and addressed by maintainers. The challenge for organisations is that patching and deployment may become more time-sensitive even as automated tools help defenders test code. For users, keeping operating systems, browsers and other software updated remains a direct way to receive available fixes.
METR’s findings add a separate concern: AI may be helping labs improve the systems that will perform future work. If that feedback loop continues, capability assessments could become outdated more quickly. But METR’s estimate about AI’s role in Anthropic’s progress should be read alongside its caveat that much of the evidence was withheld from public review.
Products Worth Considering
Jackery Explorer 300 Portable Power Station with 40W Solar Panel
DJI Osmo 360 Road Cycling First-Person Combo 360° Action Camera
Apple iPad 11-inch with A16 Chip, 512GB, Wi-Fi 6, and Apple Pencil (USB-C) — Blue
As an affiliate, we earn on qualifying purchases.
How the Two Findings Connect
The OpenAI and METR announcements concern different organisations and different measurements. OpenAI reported results from a specific Linux security campaign; METR assessed the pace of capability change using Anthropic’s internal data. Together they suggest a possible cycle in which AI helps identify software weaknesses while also helping researchers build more capable AI systems. They do not, by themselves, prove that the same tools or processes are used across the two labs.
The source account also cites Mozilla advisories crediting OpenAI with 50 high-impact Firefox fixes in October. Those public advisories do not identify which model found each issue, so they cannot independently confirm GPT-6 Cyber’s role in those discoveries. It also describes a separate Epoch AI finding that a downloadable model called GLM 5.3 could turn published Firefox fixes into working attacks for about $1,000 per Firefox version. That is a distinct result and should not be confused with OpenAI’s Linux campaign.
OpenAI’s reported campaign followed an earlier July run that the source says produced 166 Linux privilege-escalation bugs. The October results were disclosed before OpenAI named GPT-6 Cyber as the model behind them on November 5. The source says the Linux findings were shared for remediation, but the material provided does not give a public account from Linux maintainers about the findings or their status.
““The public has to rely on METR’s judgment much more than would be ideal.””
— METR, on the limits of its public evidence
What the Public Record Cannot Verify
The detailed Linux figures are OpenAI’s own account. The material provided does not include an independent technical review of all 948 exploits, the 588 claims of previously unknown vulnerabilities, or the cost calculation. Nor does it include public comment from Linux kernel maintainers on the campaign’s findings, their remediation status or how many were confirmed as distinct security issues.
It is also unclear how much the results depended on the model versus the surrounding scanning, testing and engineering pipeline. The reported API bill does not cover all potential costs, such as human review or maintenance work. The Firefox advisories cited in the source credit OpenAI with fixes but do not establish which model discovered particular bugs.
METR’s assessment is constrained by restricted access to Anthropic’s internal data. The report provides a measured acceleration and an estimate that AI accounts for about 80% of it, but much of the underlying evidence is not public. The estimate should not be treated as a confirmed causal breakdown, and the source material does not establish how quickly the trend will continue.
Disclosure, Patches and Wider Access
OpenAI says GPT-6 Cyber is available only to organisations it approves for Daybreak Red, with wider availability to approved Red customers expected over the coming weeks, according to coverage cited in the source. The company says access includes identity checks, hardware security keys, managed devices, restricted API access and usage monitoring. Public information about how those safeguards perform in practice remains limited.
The next concrete security milestones are confirmation from maintainers and affected software projects about the Linux findings, along with details of fixes and deployment. The source says the findings were sent for remediation, but it does not provide a public schedule or a count of patches already released. Readers should install security updates promptly as they become available.
For METR’s findings, further public evidence or future measurements could clarify whether the faster pace persists and how much AI contributes to researchers’ work. Until then, the report is an informed assessment with acknowledged evidence limits, while OpenAI’s exploit totals remain company-reported results.
Key Questions
What is GPT-6 Cyber?
OpenAI describes GPT-6 Cyber as a cyber-specialized version of GPT-6 Astra for authorized security work, including vulnerability discovery and testing. Access is restricted to organisations approved for its Daybreak Red programme.
Did OpenAI report 948 confirmed Linux vulnerabilities?
OpenAI reported 948 working local privilege-escalation exploits, after describing a process that reproduced 1,724 vulnerabilities in test machines and grouped exploit results by root cause. The figures have not been independently audited in the source material, and the status of individual findings is not fully public.
What does $1,279 per exploit mean?
It is the reported $1,213,430 API bill divided by 948 working exploits. It does not represent the full cost of validation, disclosure, fixing the vulnerabilities or deploying patches.
What did METR find about Anthropic?
METR said its main measure of AI capability improvement grew about 1.9 times faster after October 2025 than before. It estimated AI assistance may account for roughly 80% of that acceleration, while noting that much of its most informative evidence was redacted.
What should Linux users do now?
The source material does not identify a specific Linux patch users can install for these reported findings. Users should keep their distributions and software updated and apply relevant security updates when their maintainers release them.
Source: ThorstenMeyerAI.com
