The GPT-6 Cyber And METR Findings: AI Exploits, Speed, And Context

🔍 Read the full analysis: The GPT-6 Cyber And METR Findings: AI Exploits, Speed, And Context 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 named GPT-6 Cyber on November 5 as the model behind a Linux kernel security exercise that the company says produced 948 working privilege-escalation exploits, including 588 previously unknown flaws. On the same day, evaluator METR reported faster measured progress at Anthropic, linking much of the acceleration it estimates to AI helping researchers. The findings suggest AI is advancing both vulnerability research and AI development, though important evidence remains difficult to check independently.

OpenAI describes GPT-6 Cyber as a cyber-specialized version of GPT-6 Astra for authorized security work. Access is restricted to organizations approved for the company’s Daybreak Red programme, which covers activities such as penetration testing and red-team work. OpenAI says access requires safeguards including identity checks, multi-factor authentication, managed devices and restricted API connections. The company rates the model “Critical” under its own preparedness framework, the framework’s highest cybersecurity tier; that is OpenAI’s assessment, not an independent rating.

In a campaign conducted in mid-October, OpenAI says its system scanned about 2.8 million lines of Linux kernel code, identified 2,308 candidate findings and reproduced 1,724 vulnerabilities in test machines. It reported producing 948 working local privilege-escalation exploits after grouping findings by root cause, with 588 described as previously unknown. OpenAI’s figures are company-reported. Mozilla’s public advisories credit OpenAI with 50 high-impact Firefox fixes in October, but do not identify which model found each flaw. The source material says Linux maintainers had not publicly commented at the time of publication.

METR’s separate report says its main measure of Anthropic’s progress grew about 1.9 times faster after October 2025 than before. METR estimates that AI assistance may account for around 80% of the acceleration, while warning that much of its most informative evidence was redacted. The report is an evaluator’s analysis, not a public release of the underlying Anthropic data. It does not establish that AI alone caused the change.

Faster Exploit Discovery, Tighter Deadlines

If the reported results hold up, security teams face a shorter interval between a software flaw becoming discoverable and a usable exploit being produced. OpenAI’s comparison with an earlier Linux exercise is striking: it reported 166 privilege-escalation bugs for $513,690 in July, versus 948 exploits for about $1.21 million in October. The cost per reported exploit works out to roughly $3,100 in the earlier run and $1,279 in the later one. These are figures from OpenAI’s exercises; they are not a universal market price or an independently audited measure of security work.

Lower costs could help defenders find and fix flaws, but the same capability may also be available to attackers. The source material cites Epoch AI’s late-October report that a freely downloadable model, GLM 5.3, converted published Firefox fixes into working attacks for about $1,000 per Firefox version. That is a separate research result, not evidence that GPT-6 Cyber is openly available. OpenAI says GPT-6 Cyber is restricted to vetted organizations, but access controls cannot remove the broader risk that similar techniques spread through other models or public research.

METR’s findings matter for a different reason: if AI tools are helping researchers improve AI systems faster, the pace of future capabilities may itself increase. METR’s estimate is uncertain and relies on data that the public cannot fully inspect. Taken together, the reports make patching speed and independent evaluation central concerns for businesses, software maintainers and users.

Products Worth Considering

Samsung Galaxy Tab A7 Lite 8.7-Inch 32GB Wi-Fi Tablet, Gray

AmazonView Latest Price

Samsung Galaxy Tab A11 Wi-Fi Tablet, 8.7-Inch, 64GB, 4GB RAM, Gray (SM-X133, International Version)

AmazonView Latest Price

Samsung Galaxy Tab A11+ 11-Inch Tablet, 6GB RAM, 128GB Storage, Expandable, Gray

AmazonView Latest Price

As an affiliate, we earn on qualifying purchases.

Restricted Release, Partial Verification

GPT-6 Cyber is described as OpenAI’s fourth cyber-focused model released in 2026, following releases in April, June and August. Its Daybreak programme has separate tiers: Blue for defensive tasks such as code review and incident response, and Red for offensive security testing. GPT-6 Cyber is limited to approved Red organizations. OpenAI says it can monitor use and suspend or revoke access. The company’s listed price for approved users is $20 per million input tokens and $100 per million output tokens.

The Linux campaign figures were first described by OpenAI in a November 4 post titled “A thousand fewer paths to root”; OpenAI named GPT-6 Cyber as the model behind the work the following day. The count of 588 previously unknown flaws and the exploit totals are attributed to OpenAI in the source material. Mozilla’s advisories provide a separate, public record of fixes, but they do not connect specific fixes to the model or validate the Linux totals. The source also cites an earlier July OpenAI run, which offers a company-reported comparison over time, not a controlled independent benchmark.

METR’s report examines changes in Anthropic’s progress using data from inside the company over an eleven-week period. METR says the public must rely on its judgment more than it would prefer because Anthropic redacted a majority of the evidence it found most informative. The source article also discloses that its author used Anthropic’s Claude in drafting the analysis, while covering Anthropic and its competitor OpenAI.

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

— METR, in its report on Anthropic

What the Public Cannot Verify

The source material does not provide an independent audit of OpenAI’s Linux campaign, its methodology or the claim that 588 exploits were previously unknown. It also does not identify which of the reported findings have been disclosed, patched or publicly documented. Mozilla’s advisories confirm OpenAI’s role in 50 high-impact Firefox fixes, but do not establish which model found them or whether they relate to the Linux work. No public comment from Linux maintainers was reported in the source.

METR’s estimate that AI accounts for roughly 80% of Anthropic’s acceleration remains an estimate, with much of the evidence METR considers most informative withheld. The report does not settle how much of the measured change came from AI tools, other changes in research practice or different factors. It is also unclear how quickly GPT-6 Cyber access will expand beyond the approved organizations and how well OpenAI’s monitoring works in practice.

Disclosure, Patches and Wider Access

OpenAI says GPT-6 Cyber access is being provided to approved Daybreak Red organizations, with reports in the source material indicating rollout to eligible customers over the coming weeks. Further public detail from OpenAI, affected software maintainers and security advisories could help establish which findings were confirmed and what has been fixed. The source does not give a public schedule for release of a complete Linux vulnerability list or an independent review of the exploit counts.

For organizations, the immediate response is to monitor vendor security notices, apply available patches promptly and review how quickly vulnerability reports can be triaged. Users should install operating system and browser updates as they become available. METR’s report may prompt further scrutiny of how labs measure and disclose AI-assisted research progress; no additional release of Anthropic’s underlying data is specified in the source material.

Key Questions

What is GPT-6 Cyber?

It is OpenAI’s cyber-specialized version of GPT-6 Astra, intended for authorized security work. OpenAI says access is restricted to approved organizations in its Daybreak Red programme.

Did GPT-6 Cyber find 948 confirmed Linux flaws?

OpenAI says its mid-October campaign produced 948 working privilege-escalation exploits, including 588 it described as previously unknown. The figures have not been independently audited in the material provided, and the public status of individual Linux findings is unclear.

What did METR find about Anthropic?

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

Does the reporting show that AI is causing more cyberattacks?

No. The reports describe AI capabilities and security research results; they do not establish a rise in real-world attacks caused by these models. OpenAI says GPT-6 Cyber is restricted to vetted organizations, while the broader potential for misuse remains a concern.

What should computer users do now?

Install operating system and browser updates promptly, and follow security notices from software providers. The reports do not change the basic value of keeping software patched.

Source: ThorstenMeyerAI.com