🔍 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 privacy and security gear 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 run of agents powered by GPT-6 Cyber produced 948 working Linux privilege-escalation exploits, including 588 it says were previously unknown, at an API cost of about $1,279 per exploit. Separately, METR estimates Anthropic’s pace of progress on its measure of AI capabilities rose about 1.9 times after October 2025, while warning that much of its supporting evidence is confidential.
OpenAI said on November 5 that its restricted cybersecurity model, GPT-6 Cyber, powered a mid-October run that generated 948 working Linux privilege-escalation exploits, including 588 vulnerabilities the company says were previously unknown. On the same day, independent evaluator METR reported that Anthropic’s measured pace of AI progress had risen about 1.9 times after October 2025, with AI assistance likely accounting for much of the increase.
OpenAI’s reported Linux campaign cost $1,213,430 in API use, or about $1,279 per working exploit. The company says its pipeline examined roughly 2.8 million lines of kernel code, analyzed 2,308 candidate findings and reproduced 1,724 vulnerabilities in test machines before producing the 948 working exploits. The figures come from OpenAI’s account; the source material does not cite an independent audit of the campaign’s full results.
GPT-6 Cyber is a cybersecurity-focused version of GPT-6 Astra, designed for vulnerability discovery, exploit development and web application testing, according to OpenAI’s system card. Access is limited to organizations approved for the Daybreak Red programme, which covers offensive security testing. OpenAI says approved users face identity, device and access controls, and that it can monitor, suspend or revoke access. The company rates the model “Critical” under its own preparedness framework, its highest cybersecurity tier; that rating is not an external assessment.
METR’s separate report examined Anthropic’s internal development data over 11 weeks. It says Anthropic’s main measure of progress rose 1.9 times faster after October 2025 than before. METR estimates that AI systems may account for about 80% of the acceleration, but says much of the evidence it found most informative was redacted. Its estimate is an analysis, not a fully public audit of Anthropic’s internal work.
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 Model Development
The reports point to two pressures that may reinforce each other: AI tools are being used to identify and exploit software weaknesses, while AI assistance may be speeding the work of building more capable systems. The figures do not establish that every lab or attacker can reproduce these results, but they raise a practical concern for software maintainers: finding vulnerabilities may become cheaper and faster.
OpenAI’s reported cost is a measure of one company’s API spending in one campaign, not the full cost of validating, responsibly reporting and fixing vulnerabilities. Still, if similar capabilities become available to more users, the time between discovery and potential misuse could narrow. Security teams may have less time to patch exposed systems, making rapid updates and vulnerability response more consequential for organizations and everyday users.
The METR findings matter because faster research cycles could shorten the time between AI generations. But the report’s estimate about AI’s contribution is uncertain, and the supporting information is partly withheld. Together, the announcements offer evidence of a developing trend, not proof that all AI research is accelerating at the same rate or that the exploit results are broadly replicable.
Products Worth Considering
DJI Osmo Mobile 8 Standard Combo Phone Gimbal Stabilizer
Bose TV Speaker (2nd Gen) Compact Soundbar
Bose QuietComfort Ultra Earbuds (2nd Gen), Wireless Noise-Cancelling Earbuds, Black
As an affiliate, we earn on qualifying purchases.
From July’s Run to November’s Model
OpenAI reported a previous Linux campaign in July that found 166 privilege-escalation bugs for $513,690, about $3,100 per exploit. The October run’s reported total is larger and its reported cost per exploit lower. These are two company-reported runs, however, and the available material does not establish that their conditions were identical, so the comparison should not be treated as a controlled benchmark.
OpenAI also reported that its agents found 81 previously unknown high-severity Firefox flaws. Mozilla’s public advisories credit OpenAI with 50 high-impact Firefox fixes in October, but do not identify which model found each issue. The source material says no Linux maintainer had commented publicly at the time of publication. Separately, Epoch AI reported in late October that a downloadable model called GLM 5.3 could turn published Firefox fixes into working attacks at a cost of about $1,000 per Firefox version. That is a distinct result and should not be conflated with OpenAI’s Linux campaign.
GPT-6 Cyber is OpenAI’s fourth cyber-focused model release this year, according to the source account. Its Daybreak programme separates defensive work, such as code review and incident response, from offensive testing. The model is restricted to approved Red-tier organizations. METR, an independent evaluator, has studied AI capabilities and task-length trends; in this report, it examined evidence from Anthropic while acknowledging limits on what it could make public.
“The public has to rely on METR’s judgment much more than would be ideal.”
— METR, in its report on Anthropic
Independent Checks Remain Limited
The Linux exploit totals, the count of previously unknown flaws and the API bill are based on OpenAI’s account. The source material does not provide an independent replication of the campaign, or public comments from Linux maintainers confirming the results. It also does not show how many of the reported vulnerabilities have been fixed, disclosed to affected parties or remain exploitable.
Mozilla’s advisories document high-impact Firefox fixes and credit OpenAI, but do not link each issue to GPT-6 Cyber or establish that all 81 reported findings were publicly confirmed. METR’s estimate that AI explains about 80% of Anthropic’s acceleration is also qualified: the evaluator says much of its evidence is redacted. The exact methodology behind the estimate and the full underlying data are not available in the supplied material.
It is not yet clear whether the reported cost per exploit includes all relevant engineering and validation expenses, or how results would change on other software projects. Nor do the two announcements establish a direct connection between OpenAI’s cybersecurity results and Anthropic’s development pace.
Patches and Public Evidence
OpenAI says GPT-6 Cyber access is limited to approved Daybreak Red organizations, with further access expected to reach eligible customers over the coming weeks, according to coverage cited in the source account. The company’s announced controls and access rules will be important to watch, as will any additional technical documentation about the Linux and Firefox findings.
For maintainers and security teams, the immediate next steps are to check advisories, apply available fixes and review systems for affected software. Users should install operating-system and browser updates promptly. More public confirmation from maintainers, security advisories that connect specific findings to the reported runs, and any further detail from METR or Anthropic could clarify how much of the claims can be independently checked.
Key Questions
What is GPT-6 Cyber?
GPT-6 Cyber is OpenAI’s cybersecurity-focused version of GPT-6 Astra. OpenAI says it is intended for authorized security work, including vulnerability discovery and testing.
Did OpenAI independently verify the 948 Linux exploits?
The supplied reporting describes the figure as OpenAI’s reported result. It does not cite an independent replication or public confirmation from Linux maintainers.
What does the $1,279 figure measure?
It is the reported API cost per working exploit, calculated from OpenAI’s stated $1,213,430 bill and 948 exploits. It is not a complete estimate of all costs associated with discovering, validating or fixing the vulnerabilities.
What did METR conclude about Anthropic?
METR said Anthropic’s main measured pace of progress was about 1.9 times faster after October 2025 than before. It estimated AI may account for about 80% of that acceleration, while noting that much of the supporting evidence was confidential.
What should organizations and users do now?
Organizations should monitor software security advisories and apply relevant patches quickly. Individual users should keep operating systems and browsers up to date; the reports do not identify a single confirmed flaw that affects every user.
Source: ThorstenMeyerAI.com
