How GPT-6 Cyber And The METR Report Describe A Lab Running Twice As Fast

🔍 Read the full analysis: How GPT-6 Cyber And The METR Report Describe A Lab Running Twice As Fast on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on tech for your team

• Business-only prices and quantity discounts
• Tax-exempt purchasing
• Multiple users, one account, clear invoices

Create a free Business account →Free to register

As an affiliate, we earn on qualifying purchases.Hide

OpenAI on November 5 named GPT-6 Cyber as the model behind a mid-October Linux security run that the company says produced 948 working privilege-escalation exploits, including 588 previously unknown flaws. On the same day, independent evaluator METR reported that Anthropic’s measured AI research pace accelerated after October 2025, with AI likely contributing substantially—a pairing that points to faster vulnerability discovery and faster AI development, though key evidence remains private.

OpenAI says the Linux campaign scanned about 2.8 million lines of kernel code, around 7% of the project. The company’s account describes 2,308 candidate findings, 1,724 vulnerabilities reproduced in test machines and 948 working local exploits. OpenAI said 588 of those exploits involved flaws not previously known. The reported API bill was $1,213,430, or about $1,279 per exploit. These figures come from OpenAI’s account; the supplied material does not include an independent audit of the run or its cost.

OpenAI describes GPT-6 Cyber as a cyber-specialized version of GPT-6 Astra for authorized security work, including finding vulnerabilities, building exploits and testing web applications. Access is restricted to organisations approved for the Daybreak Red programme, which is intended for offensive testing such as penetration tests and red-team work. OpenAI’s system card lists identity, device-security and access-control requirements, and says the company can monitor, pause or revoke use. Its “Critical” cybersecurity rating is OpenAI’s own assessment under its Preparedness Framework, not an independent certification.

METR’s November 5 report examined Anthropic’s internal development data over 11 weeks. It found that its main measure of AI research progress grew 1.9 times faster after October 2025 than before. METR estimates that AI assistance may account for about 80% of the current acceleration. The estimate is not a direct measurement of every contribution: METR says Anthropic withheld much of the evidence it considered most informative, limiting what outsiders can verify.

Cheaper Exploits, Faster Research

The two reports concern separate organisations and different evidence, but together they describe a development with consequences for both software security and AI oversight. If AI systems can help find and turn flaws into working exploits at lower cost, defenders may face a shorter period between a vulnerability being discoverable and being usable. That can raise pressure on software makers to identify, disclose and patch flaws quickly.

The falling cost comparison is notable but needs careful qualification. The source material compares OpenAI’s October figure of about $1,279 per exploit with a July run it says cost roughly $3,100 per exploit. Both are company-reported campaign results, and the supplied account does not establish that the runs used identical methods or produced directly comparable findings. The comparison is a reported cost trend, not proof that every real-world exploit can now be produced at that price.

METR’s finding matters because AI helping researchers improve AI could speed up future model development. But its estimate that AI accounts for about 80% of the acceleration is an attribution based on its analysis, not a fully public record that readers can independently reproduce. For businesses and users, the practical security implication remains straightforward: timely software updates matter, while the scale and pace of any resulting increase in real-world attacks are not established by these reports.

Products Worth Considering

LG UltraGear 32GS60QC-B 32-inch Curved QHD Gaming Monitor (180Hz, 1ms, 1000R, AMD FreeSync, HDR10)

AmazonView Latest Price

Apple MacBook Air 13-inch Laptop with M5 Chip, 16GB Memory, 512GB SSD, 12MP Center Stage Camera, Touch ID, Wi-Fi 7, Midnight

AmazonView Latest Price

LG UltraGear 32GX850A-B 32-inch 4K UHD Glossy OLED Gaming Monitor

AmazonView Latest Price

As an affiliate, we earn on qualifying purchases.

Two Reports, Different Evidence

OpenAI’s announcement followed its mid-October Linux run. The supplied source says the company had described that work publicly on November 4, then identified GPT-6 Cyber as the model behind it on November 5. OpenAI has also reported earlier cyber-focused model releases this year. Its Daybreak programme separates approved defensive work from offensive testing; GPT-6 Cyber is limited to organisations OpenAI approves for the Red tier. The company lists pricing of $20 per million input tokens and $100 per million output tokens for approved organisations.

Some evidence around the security work comes from outside OpenAI, but it does not independently verify the full Linux figures. The source says Mozilla’s public advisories credited OpenAI with 50 high-impact Firefox fixes in October, without specifying which model found each flaw. It also reports that no Linux maintainer had publicly commented at the time. Separately, Epoch AI reported that a freely downloadable model, GLM 5.3, could turn published Firefox fixes into working attacks at an estimated cost of about $1,000 per Firefox version. That is a separate finding, not evidence that GPT-6 Cyber produced those attacks.

METR’s work is an external evaluation of Anthropic’s research pace, not an announcement by OpenAI or a direct comparison between the two companies. The report’s conclusions rely partly on data Anthropic provided and partly on METR’s interpretation. Its stated evidence limitations matter when reading the headline figure: the 1.9-times measure describes METR’s main research-progress measure, not a claim that every part of Anthropic’s work or every model capability improved at that rate.

“The public has to rely on METR’s judgment much more than would be ideal.”

— METR, on the limits of its evidence

What the Public Cannot Verify

The Linux exploit count, the number of previously unknown flaws and the campaign’s cost are OpenAI-reported figures. The material available here does not establish that an independent party reproduced all 948 exploits, audited the billing total or confirmed how the findings were grouped. Mozilla’s advisories provide some separate evidence of security fixes, but do not identify which model discovered each issue. Public comment from Linux maintainers was not available in the source material.

METR disclosed that Anthropic withheld a majority of the evidence it considered most informative. That leaves readers unable to independently inspect much of the basis for the report’s 1.9-times acceleration finding or its estimate that AI contributed about 80%. The supplied material also does not describe enough of the underlying measure to establish how it relates to every stage of model development.

It remains unclear whether cheaper AI-assisted exploit development will lead to more attacks, quicker fixes, or both, and how widely GPT-6 Cyber will be available beyond approved Daybreak Red organisations. The reports do not establish a direct causal link between OpenAI’s cyber work and Anthropic’s research acceleration.

Access, Patches and Further Evidence

OpenAI’s access remains limited to organisations it approves for Daybreak Red. The source material, citing reporting by CSO Online and The Register, says availability to all Red customers was expected over the coming weeks; that is a reported rollout expectation, not confirmation that access has already expanded. Further detail from OpenAI about the Linux findings, or responses from kernel maintainers, could help clarify how the results were validated and what remediation followed.

For software users and organisations, the immediate step is to apply security updates promptly and keep systems maintained. Security teams will be watching for disclosures and patches tied to vulnerabilities found during the reported work, as well as for independent information about the scope of the campaign. METR’s assessment could also become easier to evaluate if more supporting evidence is released. Until then, its acceleration estimate should be read alongside its explicit caveat about restricted access to the underlying data.

Key Questions

What did OpenAI say GPT-6 Cyber did?

OpenAI said GPT-6 Cyber powered a mid-October Linux security run that produced 948 working local privilege-escalation exploits, including 588 it described as involving previously unknown flaws. The figures have not been independently verified in the supplied material.

Who can use GPT-6 Cyber?

OpenAI says access is limited to organisations it approves for the Daybreak Red programme, which covers offensive security testing. The company lists identity, device-security and access-control requirements.

What did METR find about Anthropic?

METR reported that its main measure of Anthropic’s AI research progress grew 1.9 times faster after October 2025 than before. It estimated that AI may account for about 80% of the acceleration, while warning that much of its most informative evidence was withheld.

Are the OpenAI and METR findings independently confirmed?

Not fully. OpenAI supplied the Linux exploit and cost figures, while METR conducted an external evaluation but said it could not publish much of the underlying Anthropic evidence. The source material does not provide a public independent replication of either headline result.

What should computer users do now?

Install security updates promptly and keep operating systems and applications supported. The reports describe faster vulnerability research, but do not establish how much that has changed real-world attack rates.

Source: ThorstenMeyerAI.com