🔍 Read the full analysis: What To Make Of GPT-6 Cyber, The METR Report, And A $1,279 Exploit on ThorstenMeyerAI.com
Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
TL;DR
OpenAI says its restricted GPT-6 Cyber model powered research that produced 948 working Linux privilege-escalation exploits, including 588 flaws it says were previously unknown. METR separately estimates Anthropic’s model-improvement pace has accelerated, with AI contributing to much of the change, but says key evidence is not public.
OpenAI said on November 5 that its restricted cyber model, GPT-6 Cyber, powered security research that generated 948 working Linux privilege-escalation exploits; the company says 588 involved previously unknown flaws. On the same day, independent evaluator METR published an assessment that Anthropic’s model-improvement pace has accelerated, likely in part because AI is helping its researchers work faster. The findings point to a shift in both vulnerability discovery and AI development, though much of the supporting evidence remains company-controlled.
OpenAI’s Linux campaign took place in mid-October. The company says its system scanned about 2.8 million lines of kernel code, around 7% of the project, and examined 2,308 candidate findings. It reports reproducing 1,724 vulnerabilities in test machines and producing 948 working local privilege-escalation exploits. OpenAI says 588 were previously unknown. These figures come from the company’s account; they have not been independently verified in the supplied material.
OpenAI also reported identifying 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. OpenAI’s reported API bill for the Linux work was $1,213,430, or about $1,279 per working exploit, using the company’s total and exploit count. A July run reportedly generated 166 Linux privilege-escalation bugs for $513,690, about $3,100 each; the two company-reported figures suggest lower costs per exploit in October, but do not establish that the work was otherwise comparable.
GPT-6 Cyber is a cyber-specialized version of GPT-6 Astra, according to OpenAI’s system card, and is intended for authorized security work such as vulnerability discovery, exploit development and application testing. Access is limited to organisations approved for the company’s Daybreak Red programme. OpenAI’s own framework rates the model “Critical,” its highest cybersecurity tier; that is the company’s assessment, not an independent rating.
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 Discovery, Faster Patch Pressure
The reported cost matters because security teams may face a shorter interval between a flaw being discovered and someone developing a usable exploit. If AI tools can find and reproduce vulnerabilities at scale, organisations may need to identify affected systems and apply fixes faster. The $1,279 figure is a cost per exploit for this particular OpenAI run, not a general price for finding or exploiting software flaws.
The capability is not confined to restricted commercial systems. The source material cites late-October research from Epoch AI reporting that the 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 research claim, and does not establish how broadly the method works across software. Together, the reports raise practical concerns for software makers and users: faster discovery can help defenders, but public fixes and capable tools may also make it easier for others to develop attacks.
METR’s report adds a related concern about the pace of improvement itself. If AI is helping researchers build better AI systems, advances could compound. METR estimates that AI accounts for about 80% of Anthropic’s acceleration, but presents this as an estimate based on evidence it says is partly withheld, not a directly observable or independently reproducible measurement.
Popular right now
Philips Café Aromis Automatic Espresso Machine with LatteGo Pro Milk System
De'Longhi Rivelia Fully Automatic Espresso Machine with Milk Frother, White
Apple AirPods Pro 3 Wireless Earbuds with Active Noise Cancellation
As an affiliate, we earn on qualifying purchases.
Two Reports, Different Evidence
OpenAI introduced GPT-6 Cyber on November 5, following cyber-focused model releases in April, June and August, according to the source material. Its Daybreak programme has Blue access for defensive work and Red access for offensive testing. GPT-6 Cyber is restricted to approved Red organisations. The listed controls include single sign-on, multi-factor authentication, managed devices and least-privilege access; individual researchers must pass identity checks and use a hardware security key. OpenAI says access can be monitored, paused or revoked. The stated API prices are $20 per million input tokens and $100 per million output tokens for approved organisations.
The Linux work was described by OpenAI in a November 4 post titled “A thousand fewer paths to root”; the company named the model behind it the following day. The timing matters: the output figures were presented before the model identification. The available account says Linux findings were sent for handling, but the supplied material does not give a public response from Linux maintainers. Mozilla’s advisories offer a public record of fixes, though they do not map each fix to a particular model or discovery process.
METR’s November 5 report examined Anthropic data over an eleven-week period. It says its principal measure of Anthropic’s progress grew 1.9 times faster after October 2025 than before. METR attributes about 80% of the acceleration to AI assistance as its best estimate. The comparison and causal estimate are METR’s analysis, rather than a result independently confirmed in the source material.
“The public has to rely on METR’s judgment much more than would be ideal.”
— METR, in its report on Anthropic
How Much Can Be Independently Checked?
The Linux exploit count, the claim that 588 flaws were previously unknown, and the reported costs are OpenAI’s figures. The material supplied here does not include independent validation of the full results, a breakdown of each finding, or a public statement from Linux maintainers. Nor does it establish how many findings have been patched, how many affect deployed systems, or whether the exploit count translates into real-world attacks.
Mozilla’s advisories confirm that it credited OpenAI with 50 high-impact Firefox fixes in October, but they do not show which model found each flaw. The relationship between that figure and OpenAI’s report of 81 previously unknown high-severity Firefox issues is not explained in the available material.
METR says much of its most informative evidence was withheld by Anthropic. Its reported 1.9-times acceleration and estimate that AI accounts for about 80% of it should be read as METR’s analysis, with limited public access to the underlying evidence. The supplied account also does not detail the measure sufficiently to reproduce the calculation.
Patch Status and Further Review
The immediate security question is what happens to the reported Linux findings: whether maintainers confirm them, which receive fixes, and how quickly those patches reach users. The source material provides no timetable or public Linux maintainer response. OpenAI says its Daybreak access is controlled and that it can monitor or revoke use, but the material does not provide independent evidence on how those safeguards operate in practice.
Further public documentation could clarify how the Linux and Firefox counts were calculated, which findings were fixed, and whether outside researchers can reproduce the reported results. METR’s assessment would also be easier to evaluate if more of the underlying evidence or methods became available. Until then, organisations should treat the reports as significant claims with incomplete public verification, while users should install operating-system and browser security updates promptly.
Key Questions
What is GPT-6 Cyber?
OpenAI describes it as a cyber-specialized version of GPT-6 Astra for authorized security work. Access is limited to organisations it approves for the Daybreak Red programme.
Did GPT-6 Cyber find 948 Linux vulnerabilities?
OpenAI says its October campaign produced 948 working local privilege-escalation exploits after reproducing vulnerabilities in test machines. The reported count has not been independently verified in the source material, and an exploit count is not necessarily the same as a count of distinct underlying flaws.
What does the $1,279 figure represent?
It is the reported API bill of $1,213,430 divided by the 948 working exploits OpenAI says it produced. It describes that campaign’s reported cost per exploit, not a universal cost for AI security research.
What did METR conclude about Anthropic?
METR reported that its main measure of Anthropic’s progress grew 1.9 times faster after October 2025 than before. It estimated that AI assistance accounts for about 80% of the acceleration, while warning that much of the evidence it found informative was not public.
What should software users do now?
Install security updates for operating systems, browsers and other software promptly. The reports do not establish that the cited Linux findings are being exploited in the wild, but timely patching reduces exposure to known vulnerabilities.
Source: ThorstenMeyerAI.com
