🔍 Read the full analysis: What A $1,279 Exploit Says About AI, Alongside GPT-6 Cyber And METR on ThorstenMeyerAI.com
Get hardware and tech essentials 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 a mid-October AI-assisted Linux security run produced 948 working privilege-escalation exploits, including 588 flaws it had not previously known, at an API cost of about $1,279 per exploit. On November 5, OpenAI named GPT-6 Cyber as the model behind the work, while METR reported that Anthropic’s model-development pace has accelerated and estimated AI accounts for much of that change. Important details rely on company claims, and METR said Anthropic withheld much of the evidence it reviewed.
OpenAI said on November 5 that its restricted cybersecurity model, GPT-6 Cyber, powered a mid-October run that produced 948 working Linux privilege-escalation exploits, at an API cost of about $1,279 each. The same day, independent evaluator METR reported that Anthropic’s pace of AI-model improvement has increased, linking much of the acceleration to AI assistance in research. Together, the reports point to AI contributing both to vulnerability discovery and to the development of more capable models, though key evidence remains difficult for the public to verify.
OpenAI’s account says the Linux project examined 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 exploits after grouping findings by root cause. OpenAI says 588 of the 948 flaws were previously unknown. The stated API bill was $1,213,430, or about $1,279 per working exploit. These figures come from OpenAI’s description of its own run, rather than an independently published audit of every finding and cost.
OpenAI describes GPT-6 Cyber as a cybersecurity-focused version of GPT-6 Astra, trained for vulnerability discovery, exploit development and web-application testing. Access is limited to organisations approved for the Daybreak Red programme, which is intended for authorized offensive security work such as penetration testing and red teaming. OpenAI says applicants must meet security requirements including multifactor authentication, managed devices and restricted API access. The company rates the model “Critical” for cybersecurity under its own Preparedness Framework; that is OpenAI’s assessment, not an outside certification.
METR’s separate report examined Anthropic’s internal data over an eleven-week period. It found that the organization’s main measure of model-development progress grew about 1.9 times faster after October 2025 than before. METR estimated that AI assistance may account for about 80% of the acceleration. It also said Anthropic redacted a majority of the evidence METR considered most informative, limiting what outsiders can check. The estimate is METR’s interpretation of the data it reviewed, not a publicly reproducible measurement based on a complete dataset.
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
Lower Costs Put Pressure on Patch Timelines
The reported cost matters because it suggests that producing working exploits may require less spending than earlier examples. OpenAI’s source material compares its October Linux run with a July run that produced 166 privilege-escalation bugs for $513,690, or about $3,100 each. Those figures are not a universal price for exploit development, and the runs may not be directly comparable. Still, if similar capabilities become broadly available, security teams may have less time between a vulnerability being discovered and attackers trying to exploit it.
The concern is not limited to restricted commercial models. The source article cites Epoch AI research reporting that a freely downloadable model, GLM 5.3, could turn published Firefox fixes into working attacks for about $1,000 per Firefox version. That result is also attributed to the researchers, and does not establish that every software flaw can be exploited this way. It does underline that defenders may not retain exclusive access to AI-assisted vulnerability work.
For organizations, the practical consequence is a need to identify affected systems and apply verified security updates promptly. For individual users, the clearest guidance in the source material is similarly direct: install software updates when they become available. The reported findings do not show that ordinary users are currently being targeted through these specific Linux flaws.
Products Worth Considering
Trezor Safe 7 Crypto Hardware Wallet
Ledger Flex + Folio Crypto Hardware Wallet
Trezor Safe 5 Crypto Hardware Wallet with Color Touchscreen, Black Graphite
As an affiliate, we earn on qualifying purchases.
From Linux Testing to Restricted Release
OpenAI described its Linux campaign in a November 4 post titled “A thousand fewer paths to root.” A local privilege-escalation exploit can allow someone who already has an account or code execution on a Linux machine to gain administrator-level control. That is different from an attack that starts remotely with no access to the system. OpenAI later identified GPT-6 Cyber as the model used in the work.
The source material also points to Mozilla security advisories crediting OpenAI with 50 high-impact Firefox fixes in October. Those advisories do not identify which model discovered each issue, so they cannot independently verify GPT-6 Cyber’s contribution. The source says OpenAI’s reported count of 81 previously unknown high-severity Firefox flaws is its own figure. It also reports that no Linux kernel maintainer had publicly commented on the campaign at the time of publication.
GPT-6 Cyber is described as OpenAI’s fourth cyber-focused model release of 2026, following releases in April, June and August. The model is restricted to approved Daybreak Red customers, with access through the API and Codex. OpenAI lists pricing of $20 per million input tokens and $100 per million output tokens for approved organizations. Those access controls may limit who can use the system directly, but they do not resolve the broader question of how quickly similar capabilities could spread through other models.
““the public has to rely on METR’s judgment much more than would be ideal.””
— METR, on access to Anthropic’s evidence
How Much Can Be Independently Checked
OpenAI’s Linux totals have not been independently audited in the source material. It remains unclear which vulnerabilities were disclosed to maintainers, which have been patched, and whether all 948 exploits were tested under conditions comparable to real-world systems. The source reports that 588 flaws were previously unknown, but does not provide a public, itemized list that would let readers verify that count or examine the status of each flaw.
The Firefox figures also require care: Mozilla’s advisories credit OpenAI with 50 high-impact fixes in October, but do not specify which model found each bug or confirm the separate claim of 81 previously unknown high-severity flaws. No public comment from a Linux kernel maintainer is included in the supplied material. It is not clear from these reports whether any of the described Linux flaws have been exploited outside testing.
METR’s conclusions face a different limitation. The evaluator had access to Anthropic data during its review, but said the company withheld a majority of the evidence it found most informative. The report’s 1.9-times comparison and estimate that AI explains about 80% of the acceleration therefore cannot be fully reproduced from the public material described here. The period and measure are specific to METR’s analysis; they should not be treated as a forecast of future model performance.
Disclosure, Patches and Broader Access
OpenAI’s Daybreak Red rollout is expected to reach approved customers over the coming weeks, according to coverage cited in the source material. The company says it can monitor use and pause or revoke access. Further public detail on the Linux findings, including disclosure and patch status, would help maintainers and users understand the practical risk. No public timetable for such disclosures is given in the source material.
Security teams should watch for advisories from Linux distributors and software maintainers and apply relevant updates after checking their systems. The reports do not establish that every machine is affected, so organizations will need to rely on specific vulnerability notices rather than the headline totals alone. Users should keep operating systems and applications up to date.
For METR’s findings, more public evidence or a later follow-up could clarify how its progress measure was calculated and how much confidence to place in the estimate about AI’s role. Until then, the confirmed public development is that METR reported faster progress based on data it reviewed, while acknowledging that much of the supporting evidence was not made available to readers.
Key Questions
What did OpenAI say GPT-6 Cyber did?
OpenAI said GPT-6 Cyber powered a mid-October Linux security run that generated 948 working local privilege-escalation exploits. The company said 588 related flaws had not been known before. The counts have not been independently audited in the source material.
What does $1,279 per exploit mean?
It is the reported API bill of $1,213,430 divided by the 948 working exploits in OpenAI’s account. It describes the cost of that particular run, not a guaranteed or universal cost for producing an exploit.
Can anyone use GPT-6 Cyber?
No. OpenAI says access is limited to organizations it approves for the Daybreak Red programme, which covers authorized offensive security work. The source says OpenAI planned to expand availability to approved Red customers over the coming weeks.
What did METR report about Anthropic?
METR reported that Anthropic’s main measure of model-development progress grew about 1.9 times faster after October 2025 than before. METR estimated AI assistance accounted for about 80% of the acceleration, while warning that much of the evidence it reviewed was redacted.
Do the reports show that these Linux flaws are being exploited?
No such conclusion is established by the supplied material. It describes exploits produced in testing and says public information about Linux maintainer responses and patch status was limited. Users should follow relevant security advisories and install updates when available.
Source: ThorstenMeyerAI.com
