🔍 Read the full analysis: What The $1,279 Exploit And Twice-as-Fast Lab Reveal About AI 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 GPT-6 Cyber helped produce 948 working Linux privilege-escalation exploits in an October test, at an average API cost of about $1,279 each; 588 were previously unknown, according to OpenAI. Separately, METR reports that Anthropic’s measured model-improvement pace increased about 1.9 times after October 2025, and estimates AI may account for roughly 80% of the acceleration. Key details depend heavily on company data, and METR says much of its most informative evidence was redacted.
OpenAI says its restricted GPT-6 Cyber model helped produce 948 working Linux privilege-escalation exploits during a mid-October test, at an average API cost of about $1,279 per exploit. On November 5, independent evaluator METR separately reported that Anthropic’s measured pace of model improvement had accelerated to about 1.9 times its earlier rate, with AI itself possibly driving much of the increase. Together, the findings point to AI tools becoming more capable in cybersecurity and potentially helping AI labs advance their own systems faster, though much of the evidence remains difficult to verify independently.
OpenAI’s November 4 account of the Linux exercise says the process scanned about 2.8 million lines of kernel code, or roughly 7% of the project. It identified 2,308 candidate findings, reproduced 1,724 vulnerabilities in test machines and generated 948 working local privilege-escalation exploits, according to the company’s figures. OpenAI says 588 of the exploits involved previously unknown flaws. The reported $1,213,430 API bill divided by 948 comes to about $1,279 per exploit; it is a cost estimate for this run, not a measure of total security costs.
OpenAI named GPT-6 Cyber as the model behind the work on November 5. The company describes it as a cyber-specialized version of GPT-6 Astra, intended for authorized security tasks such as vulnerability discovery and exploit testing. Access is limited to organizations approved for the offensive-testing tier of OpenAI’s Daybreak programme. The company says it requires safeguards including identity checks, hardware security keys and restrictions on API access. OpenAI rates the model “Critical” under its own cybersecurity preparedness framework; that is the company’s classification, not an independent rating.
The other development came from METR, which said its analysis of eleven weeks of Anthropic data found that its main measure of model-improvement pace rose about 1.9 times after October 2025 compared with the earlier period. METR estimates that AI assistance may explain roughly 80% of the acceleration. That figure is an estimate, not a directly observed share. METR says Anthropic withheld a majority of the evidence it considered most informative, limiting what outside readers can check.
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 Shrink Patch Windows
If OpenAI’s figures reflect capability that can be reproduced beyond its test environment, the cost of finding and developing software exploits may fall substantially. That matters to security teams and software maintainers: vulnerabilities may be found faster, but attackers could also use similar techniques to identify weaknesses or turn disclosed flaws into attacks. The reported cost covers API use in one exercise and does not establish the full cost, reliability or effort required to find and validate flaws.
The defensive value depends on how quickly organizations can verify findings and deploy fixes. Automated discovery can give maintainers more reports to triage, while a shorter gap between finding a flaw and exploiting it could leave users exposed for less time only if patches are prepared and installed promptly. The source material says Mozilla advisories credited OpenAI with 50 high-impact Firefox fixes in October, but does not identify which model found each issue. It also reports that no Linux maintainer had commented publicly at the time of publication.
METR’s findings add a longer-term concern: AI may be contributing to the research process that improves future AI systems. If that estimate is broadly right, the pace of capability change could be influenced by feedback between AI tools and the teams building them. The available report does not establish that this acceleration will continue or show that it applies equally across companies.
Products Worth Considering
Trezor Safe 5 Crypto Hardware Wallet with Color Touchscreen, Black Graphite
LG UltraGear 45GX900A-B 45-Inch Curved OLED Gaming Monitor
Trezor Safe 7 Crypto Hardware Wallet
As an affiliate, we earn on qualifying purchases.
From Linux Tests to Lab Pace
OpenAI’s Linux result follows an earlier run described in the source material. In July, the company reportedly found 166 Linux privilege-escalation bugs at a cost of $513,690, or about $3,100 per reported exploit. The October figures are larger and the reported cost per exploit is lower, but differences in methods, scope and validation mean the figures alone do not prove a like-for-like productivity gain.
OpenAI’s public account says the Linux work involved screening code, reproducing candidate vulnerabilities and generating exploits in test machines. Its reported count of 948 is lower than the 1,169 exploits mentioned before grouping findings by root cause. The distinction matters: exploit outputs, underlying vulnerabilities and distinct root causes are not interchangeable counts. The source also cites Mozilla advisories crediting OpenAI with 50 high-impact Firefox fixes during October, while separately reporting OpenAI’s claim of 81 previously unknown high-severity Firefox flaws. The advisories do not identify the model responsible for each fix.
METR’s report is a separate evaluation, not a finding about GPT-6 Cyber. Its measure tracks the pace at which Anthropic’s models improve on tasks, and its estimate about AI’s role concerns Anthropic’s research process. The two announcements therefore describe related but distinct developments: cybersecurity capabilities on one side and the use of AI in AI research on the other.
“The public has to rely on METR’s judgment much more than would be ideal.”
— METR, on the limits of its review
What Independent Checks Can Establish
The Linux totals and the API bill are OpenAI’s reported figures. The source material does not provide an independent audit of the full run, its test conditions or the calculation behind the average cost per exploit. It is also not clear how many findings were accepted by Linux maintainers, how they were disclosed, or whether the 588 described as previously unknown were all confirmed as distinct vulnerabilities.
Public Mozilla advisories provide a separate record of fixes credited to OpenAI, but they do not connect each fix to GPT-6 Cyber or verify the company’s separate claim of 81 previously unknown high-severity Firefox flaws. No public comment from a Linux kernel maintainer was reported in the source material. The reported comparison with July also lacks enough methodological detail here to establish that both runs used identical scope and procedures.
METR’s assessment has a different limitation: the organization says Anthropic redacted a majority of the evidence it considered most informative. Its estimate that AI accounts for about 80% of the acceleration should be treated as an attributed estimate, not a settled causal finding. The material also does not establish whether the measured pace will persist or how it compares with other labs.
Verification and Patch Response
OpenAI says GPT-6 Cyber is being made available through its API and Codex only to organizations it approves for the Daybreak Red tier. The source material says access for all eligible Red customers was expected to expand over the following weeks; the exact rollout status beyond that expectation is not confirmed here. Further public information about the model’s performance, controls and security findings would help clarify how its capabilities work outside OpenAI’s reported test.
For the Linux findings, the next useful markers are maintainer responses, confirmed vulnerability records and disclosures that explain which issues have been fixed. For Firefox, advisories can document affected versions and patches, though they do not by themselves identify which model discovered a flaw. Organizations should follow vendor security notices and install updates promptly; the article’s reported figures do not show that every system is affected.
METR’s estimates may be easier to evaluate if more of the underlying evidence or methodology becomes public, or if later evaluations reproduce the measured acceleration. Until then, the two developments establish a serious reported capability and a measured trend with important evidentiary limits—not a complete account of AI’s effect on software security or research.
Key Questions
What does a Linux privilege-escalation exploit do?
It can allow a user with ordinary access on a Linux machine to gain higher privileges, potentially reaching administrator-level control. OpenAI says its reported exploits were produced and tested in controlled machines.Is $1,279 the total cost of finding a vulnerability?
No. It is the approximate API bill per working exploit, calculated from OpenAI’s reported $1,213,430 bill and 948 exploits. It does not include every possible research, testing, disclosure or remediation cost.Did independent researchers verify all 948 exploits?
The source material does not report an independent audit of the full Linux run. The count and the number described as previously unknown are attributed to OpenAI.What did METR say about Anthropic’s AI research?
METR reported that its main measure of Anthropic’s model-improvement pace was about 1.9 times higher after October 2025 than before. It estimated AI may account for roughly 80% of the acceleration, while acknowledging that much of its most informative evidence was redacted.What should computer users do now?
Install security updates from software and operating-system providers promptly, and follow relevant security advisories. The reported findings do not establish that every Linux or Firefox installation is affected.Source: ThorstenMeyerAI.com
