🔍 Read the full analysis: What AI Readers Should Know About GPT-6 Cyber And The METR Report 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 named GPT-6 Cyber on Nov. 5 and attributed large-scale Linux and Firefox security findings to its cyber-focused model, while restricting access to vetted organizations. METR separately reported that Anthropic’s measured AI progress accelerated after October 2025, but said much of the supporting evidence is redacted. The cybersecurity figures largely come from OpenAI, and the METR finding carries substantial evidence limits.
OpenAI named GPT-6 Cyber on Nov. 5, describing it as a restricted model for authorized security work and identifying it as the system behind a reported campaign that produced 948 working Linux privilege-escalation exploits. That same day, independent evaluator METR reported that Anthropic’s measured model progress has accelerated since October 2025, while warning that much of the evidence it reviewed is not public.
OpenAI says 588 of the 948 Linux exploits involved previously unknown flaws. The company’s account describes a mid-October effort that scanned about 2.8 million lines of kernel code, examined 2,308 candidate findings and reproduced 1,724 vulnerabilities in test machines. The resulting exploits could let an ordinary local user gain administrator-level access. OpenAI put the API cost at $1,213,430, or about $1,279 per working exploit. Those figures are company-reported, not an independently published audit of the full run.
OpenAI also reported that its work identified 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 bug. OpenAI has described GPT-6 Cyber as a cyber-specialized variant of GPT-6 Astra, intended for vulnerability discovery, exploit development and web application testing. It rates the model “Critical” under its own cybersecurity preparedness framework; that is OpenAI’s assessment, not an outside certification.
Access is limited to organizations OpenAI approves for the Daybreak Red programme, which covers offensive testing such as penetration testing and red teaming. OpenAI’s system card lists controls including multifactor authentication, managed devices and restricted API access. The company says it monitors use and can suspend or revoke access. METR’s separate report says Anthropic’s main measure of model progress grew about 1.9 times faster after October 2025 than before. METR estimates AI assistance may account for roughly 80% of that acceleration, but stresses that the estimate relies on evidence it cannot fully disclose.
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 Tighten Patch Timelines
If OpenAI’s reported results are representative, AI-assisted vulnerability research could lower the cost and time needed to find exploitable software weaknesses. The source compares the October Linux run 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 comparison is based on the company’s reported figures, and does not by itself establish how the results would generalize to other systems or teams.
The security implications cut both ways: defenders can use models to find and fix flaws, while attackers may use similar capabilities to identify weaknesses. The source also cites Epoch AI’s late-October demonstration that a downloadable model, GLM 5.3, could turn published Firefox fixes into working attacks for about $1,000 per Firefox version. That result is a separate example, not evidence that GPT-6 Cyber was used in an attack. For organizations, the practical pressure is to review security reports and deploy patches quickly; for users, installing software updates promptly remains a direct way to reduce exposure.
METR’s finding matters for a different reason. If AI tools are helping researchers build better AI systems faster, capability changes may come more quickly than conventional development timelines suggest. The report presents this as an estimate about Anthropic’s research process, not a definitive measurement of how all AI labs are progressing.
Products Worth Considering
Dell XPS 13 9350 Copilot+ Laptop, 13.4-inch 120Hz Display, Intel Core Ultra 7 256V, 16GB RAM, 512GB SSD, Ubuntu Linux 24.04 LTS
FLEXISPOT 36-Inch Standing Desk Converter, Black
Logitech MX Master 4 for Mac Wireless Ergonomic Mouse, Space Black
As an affiliate, we earn on qualifying purchases.
Two Reports, Different Evidence
The two developments should not be treated as one study. OpenAI’s cybersecurity numbers describe work it says its own models performed. Public Mozilla advisories provide some corroboration that OpenAI was credited with high-impact Firefox fixes, but do not connect each fix to GPT-6 Cyber. The source says no Linux kernel maintainer had commented publicly at the time of writing, leaving the Linux findings without that form of outside confirmation.
METR’s report examines Anthropic’s rate of progress using the evaluator’s own measure, drawing on access to company data over eleven weeks. METR says the measure increased about 1.9 times faster after October 2025 than before, and attributes an estimated 80% of the acceleration to AI helping Anthropic’s researchers. It also says Anthropic redacted a majority of the evidence METR found most informative. That limitation is central to interpreting the report: readers can assess METR’s conclusion and stated method, but cannot independently inspect much of the underlying material.
The source places both developments within a wider shift toward AI systems assisting technical work, from security analysis to AI research. It also notes that OpenAI has released several cyber-focused models during 2026. The reports do not establish that the two companies’ findings are directly comparable, or that the same rate of progress applies across the industry.
“the public has to rely on METR’s judgment much more than would be ideal”
— METR, on the limits of its public evidence
Independent Checks Remain Limited
The full basis for OpenAI’s Linux totals has not been independently made public in the supplied material. It is not clear whether Linux maintainers have reviewed or confirmed the reported findings, or how many vulnerabilities have been patched and disclosed. Mozilla’s advisories confirm credits for high-impact Firefox fixes, but do not establish that GPT-6 Cyber found each one or substantiate OpenAI’s figure of 81 previously unknown flaws.
METR’s report also has a significant verification limit: the evaluator says Anthropic withheld a majority of the evidence it considered most informative. Its estimate that AI contributed about 80% of the acceleration is an estimate, not a directly observable share of research output. The material does not provide enough public evidence to determine whether that rate will continue or apply to other labs.
OpenAI’s restricted-access controls may limit misuse by approved customers, but the source does not establish how effective those safeguards will be in practice. It also remains unclear how quickly organizations can assess and patch vulnerabilities as automated discovery becomes less costly.
Disclosure, Access and Patching
OpenAI said GPT-6 Cyber would be made available to approved Daybreak Red organizations, with access expected to extend across eligible customers over the coming weeks. The next public markers will include further vulnerability disclosures, information about fixes and any responses from Linux maintainers. OpenAI’s system card and later technical reporting may provide more detail, but the supplied material does not specify a date for additional releases.
For Anthropic’s findings, greater visibility would depend on the company allowing more of the underlying evidence to be disclosed or on METR publishing additional method details. Until then, readers should distinguish the report’s measured trend from its estimate about AI’s contribution. Organizations should follow vendor security advisories, prioritize updates for affected systems and treat the reported figures as an indication of potential capability, not proof that every environment faces the same risk.
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 finding vulnerabilities and testing applications. Access is limited to organizations approved for its Daybreak Red programme.
Did GPT-6 Cyber produce 948 Linux exploits?
OpenAI says it did, reporting 948 working local privilege-escalation exploits from a mid-October Linux kernel campaign, including 588 it described as previously unknown. The full results have not been independently verified in the supplied material.
What did METR report about Anthropic?
METR reported that Anthropic’s main measure of model progress grew about 1.9 times faster after October 2025 than before. It estimated that AI assistance may explain roughly 80% of the acceleration, while noting that much of its underlying evidence is confidential.
Does the report prove AI systems are being used for attacks?
No. The material describes reported security research and a separate demonstration involving a downloadable model, but does not establish that GPT-6 Cyber was used in a real-world attack.
What should software users do now?
Install operating system and application security updates promptly, and follow guidance from software vendors. The reports point to faster vulnerability discovery as a possible pressure on patch timelines; they do not show that every user or system is affected.
Source: ThorstenMeyerAI.com
