🔍 Read the full analysis: Two AI Findings To Watch: A $1,279 Exploit And A Faster Lab 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 run produced 948 working Linux privilege-escalation exploits, including 588 it described as previously unknown, at an API cost of about $1,279 per exploit. Separately, METR reports that Anthropic’s measured AI-research progress accelerated after October 2025, with much of the supporting evidence withheld from public review.
OpenAI said on November 5 that its restricted cybersecurity model, GPT-6 Cyber, powered a mid-October campaign that produced 948 working Linux privilege-escalation exploits, including 588 flaws the company said were previously unknown. On the same day, independent evaluator METR reported that Anthropic’s measured AI-research progress has accelerated since October 2025, a development it attributes largely to AI assistance, though much of the evidence is not public.
OpenAI’s figures describe a process that scanned about 2.8 million lines of Linux kernel code, roughly 7% of the project, according to the company’s account. The system identified 2,308 candidate findings, reproduced 1,724 vulnerabilities in test machines and generated 948 working local privilege-escalation exploits. OpenAI said 588 of those exploits involved previously unknown flaws. The reported API bill was $1,213,430, or about $1,279 per working exploit. These are company-reported results, not an independently audited count.
The company named GPT-6 Cyber as the model behind the run on November 5. OpenAI describes it as a cybersecurity-focused version of GPT-6 Astra for authorized security work, including vulnerability research and exploit testing. Access is limited to organisations approved for the Daybreak Red programme, which covers offensive security testing. OpenAI’s system card lists security requirements such as multifactor authentication, managed devices and restricted API access. The model has OpenAI’s highest internal cybersecurity rating, “Critical”; that label reflects the company’s own framework, not an external certification.
METR’s separate assessment examined Anthropic data over eleven weeks. The evaluator said its main measure of Anthropic’s research progress grew about 1.9 times faster after October 2025 than before. METR estimated that AI systems probably account for roughly 80% of the acceleration by helping Anthropic researchers. The estimate is not a direct measurement of every source of productivity, and METR said Anthropic withheld much of the evidence it considered most informative.
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 Linux results are reproducible, they suggest that capable AI systems can help automate a substantial amount of vulnerability research and exploit development. The reported $1,279 cost per exploit is a measure of API spending for this particular campaign, not the full cost of reproducing it: it does not by itself account for research time, infrastructure, validation or responsible disclosure. Still, the figure points to a lower cost for generating working exploits than the sums cited for some past hacking competitions.
The security implications run in both directions. Automated discovery can help software maintainers find and fix flaws, but exploit techniques can also be misused. A further concern is timing: if working attacks become easier to produce, organisations may have less time to apply fixes before vulnerabilities are used against them. For users, the practical takeaway remains to install security updates promptly and use supported software.
METR’s finding matters because AI may be accelerating not only security research but also the development of future AI systems. If AI tools help researchers build and improve models faster, capabilities can advance more quickly than a lab’s earlier pace would suggest. METR’s estimate is uncertain, however, and its reliance on restricted evidence limits what outsiders can independently verify.
Products Worth Considering
Trezor Safe 7 Crypto Hardware Wallet
Trezor Safe 5 Crypto Hardware Wallet with Color Touchscreen, Black Graphite
Ledger Flex + Folio Crypto Hardware Wallet
As an affiliate, we earn on qualifying purchases.
Two Findings, Different Evidence
The OpenAI announcement follows an earlier Linux campaign described in the source material. In July, OpenAI reported finding 166 Linux privilege-escalation bugs for $513,690, or about $3,100 per bug. The October figures indicate a lower reported API cost per exploit, but comparisons should be treated cautiously: the campaigns may differ in scope, methods and what counted as a working exploit.
There is some public corroboration of OpenAI’s related Firefox security work, but not a complete independent check of the Linux claims. Mozilla advisories credited OpenAI with 50 high-impact Firefox fixes in October, according to the source report. Those advisories did not identify which model found each issue. The source also cites research by Epoch AI indicating that a 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 result and should not be conflated with OpenAI’s Linux campaign.
The two November 5 developments are therefore related as signals of AI’s growing role in technical work, but they rely on different evidence. OpenAI’s exploit totals and costs come from the company’s own account; METR’s acceleration estimate comes from an outside evaluator reviewing Anthropic data, much of which remained confidential.
“the public has to rely on METR’s judgment much more than would be ideal.”
— METR, describing the limits of its public evidence
Public Evidence Leaves Key Gaps
OpenAI’s reported exploit count, novelty tally and API cost have not been independently verified in the information available here. The public record also does not establish which of the 588 reported new findings were disclosed to maintainers, how many have been fixed, or whether any were exploited outside testing. The source report says no Linux maintainer had commented publicly at the time it was written.
Mozilla’s advisories provide evidence of credited Firefox fixes but do not map specific vulnerabilities to GPT-6 Cyber or confirm the Linux figures. OpenAI’s “Critical” label is an internal rating, and the safeguards described for Daybreak are company policies rather than independent proof that misuse cannot occur.
METR’s estimate that AI accounts for about 80% of Anthropic’s acceleration is a judgment based on limited access to evidence, not a fully public audit. The available material does not specify how much of the measured change reflects AI assistance versus other factors, or how the estimate might shift with further data.
Disclosure and Independent Checks
OpenAI says it routes security findings through authorized channels, but the supplied material does not give a public, issue-by-issue account of Linux disclosure or fixes. Further detail from OpenAI or Linux maintainers could clarify how the campaign’s findings were handled and how many vulnerabilities were remediated. OpenAI has said access to GPT-6 Cyber will be extended to approved Daybreak Red customers over the coming weeks, subject to its vetting and usage controls.
For the METR assessment, the next useful development would be additional evidence that allows outside researchers to test its conclusions while protecting confidential information. Until then, the reported acceleration should be read as an evaluator’s evidence-based estimate with acknowledged limits, rather than a fully transparent accounting of Anthropic’s research process.
For software operators and individual users, there is no need to wait for those checks before acting on routine security guidance: apply vendor updates quickly, maintain supported systems and use established security procedures. The reports describe changing research capabilities; they do not establish that a particular system is currently under attack.
Key Questions
What did OpenAI say GPT-6 Cyber produced?
OpenAI said the model powered a mid-October campaign that generated 948 working Linux privilege-escalation exploits, including 588 it described as previously unknown. These totals are based on the company’s account.
How was the $1,279-per-exploit figure calculated?
It divides OpenAI’s reported $1,213,430 API bill by 948 working exploits, yielding about $1,279 each. It is an API-cost figure for that campaign, not a full accounting of research, testing or infrastructure costs.
Can any organisation use GPT-6 Cyber?
No. OpenAI says access is restricted to organisations it approves for its Daybreak Red programme, which is intended for authorized offensive security testing. The company lists identity, device and access-control requirements.
What did METR conclude about Anthropic’s research pace?
METR reported that its main measure of Anthropic’s AI-research progress grew about 1.9 times faster after October 2025 than before. It estimated that AI assistance explains roughly 80% of the acceleration, while noting that much of its supporting evidence was withheld.
Do the reports show that Linux systems are being attacked?
No. The reported campaign involved generating and testing exploits; the material does not establish that the findings were used in real-world attacks. The status of individual disclosures and fixes is not fully clear from the available information.
Source: ThorstenMeyerAI.com
