The METR Report Says The Lab Is Running Twice As Fast—What About GPT-6 Cyber?

🔍 Read the full analysis: The METR Report Says The Lab Is Running Twice As Fast—What About GPT-6 Cyber? on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get the latest gadgets delivered free with Prime

• Fast, free delivery on millions of items
• Prime Video, Amazon Music and more included
• Member-only deals all year

Start your free Prime trial →Free trial for eligible customers · Cancel anytime

As an affiliate, we earn on qualifying purchases.Hide

OpenAI announced GPT-6 Cyber on Nov. 5, identifying it as the model behind an earlier test that the company says generated 948 working Linux privilege-escalation exploits. On the same day, independent evaluator METR reported that Anthropic’s measured AI progress has accelerated since October 2025, linking much of the change to AI assistance in the company’s own research. The claims point to faster security research and faster model development, but important parts of the evidence remain private.

OpenAI describes GPT-6 Cyber as a cyber-specialized version of GPT-6 Astra, intended for authorized security work such as vulnerability discovery, exploit development and web-application testing. Access is limited to organisations approved for the Daybreak Red programme, which covers offensive security testing. OpenAI says approved users face identity and access controls, usage monitoring, and possible suspension or revocation. The model’s “Critical” cybersecurity rating is OpenAI’s own assessment under its Preparedness Framework, not an independent evaluation.

The Linux results came from a campaign conducted in mid-October and described by OpenAI on Nov. 4. The company says the process examined about 2.8 million lines of kernel code, investigated 2,308 candidate findings and reproduced 1,724 vulnerabilities in test machines. It reports that the work yielded 948 working local privilege-escalation exploits, including 588 flaws it said were previously unknown. These figures describe OpenAI’s account of the run; the supplied material does not establish that an outside party independently reproduced the full results.

METR’s report concerns a separate development: the pace at which Anthropic’s models achieve performance on the evaluator’s task-based measure. After an 11-week review of Anthropic data, METR said its main measure grew 1.9 times faster after October 2025 than before. METR estimated that AI assistance to Anthropic researchers may account for about 80% of the acceleration. That estimate is the evaluator’s interpretation, not a directly observed breakdown of every cause.

Cheaper Discovery Pressures Patch Teams

If OpenAI’s reported Linux results hold up, they suggest that AI systems can assist with producing working exploits at a cost that is low relative to traditional vulnerability-research incentives. OpenAI’s figures put the October run at about $1,279 per exploit. The source material compares that with a July OpenAI run that reportedly produced 166 Linux privilege-escalation bugs for $513,690, or roughly $3,100 apiece. The figures are company-reported and may not represent the full cost of research, validation or remediation.

The immediate operational concern is the time between a vulnerability becoming usable and a fix reaching affected systems. Security teams may face more findings to assess, while attackers could benefit if exploit methods or vulnerability details spread beyond vetted programmes. The source also cites Epoch AI research on a freely downloadable model turning published Firefox fixes into working attacks for about $1,000 per Firefox version, suggesting the issue is not limited to restricted commercial systems. That is a separate research result, not evidence that GPT-6 Cyber was publicly available.

METR’s finding matters for a different reason: if AI is helping researchers improve AI systems more quickly, development timelines and the pace of capability changes could shift. But the report does not establish that every lab is accelerating at the same rate, or that the estimated contribution from AI applies beyond Anthropic. For organisations and users, the practical consequence remains familiar: security updates may become more time-sensitive as discovery tools improve.

Products Worth Considering

Sony WH-1000XM4C (M4 2nd Gen) Wireless Noise-Cancelling Headphones, Black

AmazonView Latest Price

Amazon Kindle Paperwhite (2024), 16 GB, 7-inch Display, Black

AmazonView Latest Price

Bose QuietComfort Ultra Earbuds (2nd Gen), Wireless Noise-Cancelling Earbuds, Black

AmazonView Latest Price

As an affiliate, we earn on qualifying purchases.

Two Reports, Different Evidence

The two announcements concern related but distinct claims. OpenAI’s report describes a particular security exercise and the company’s model behind it. METR’s report examines Anthropic’s internal progress using its own measure. The overlap is the proposed role of AI in accelerating technical work, but one set of results should not be treated as confirmation of the other.

OpenAI says GPT-6 Cyber is its fourth cyber-focused model released in 2026, following releases in April, June and August. The Daybreak programme separates defensive work from offensive testing: Blue is intended for defensive tasks, while Red covers activities such as penetration testing and red teaming. GPT-6 Cyber is restricted to approved Red organisations. OpenAI lists pricing for approved users at $20 per million input tokens and $100 per million output tokens.

There is some outside documentation related to the reported security work. Mozilla’s public advisories credit OpenAI with 50 high-impact Firefox fixes in October, according to the source material. Those advisories do not identify which model found which vulnerability. The supplied account also says no Linux maintainer had publicly commented on the Linux campaign at publication time. METR, meanwhile, cautioned that Anthropic redacted a majority of the evidence it considered most informative, leaving readers to rely more heavily on its judgment.

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

— METR, on the limits of its public evidence

Independent Checks Remain Limited

The key Linux totals—including the 948 working exploits and 588 previously unknown flaws—are attributed to OpenAI in the available material. It does not say whether an independent team verified the exploit count, the novelty of each flaw, or the full cost of the campaign. Nor does it identify how many reported findings have been patched, how the results were disclosed to maintainers, or whether the full set of technical details will be made public.

Mozilla’s advisories offer a separate, partial point of reference, but they do not connect individual Firefox fixes to GPT-6 Cyber. The source says Linux maintainers had not commented publicly, so their assessment of the campaign and its disclosure process is unknown. The reported Firefox work and the Linux campaign should not be assumed to share a methodology or verification process.

METR’s estimate that AI accounts for about 80% of Anthropic’s acceleration is also uncertain. METR says much of the underlying evidence is redacted, and its conclusion depends on a measure of task performance rather than a direct accounting of all research inputs. The reports do not establish whether this pace will continue, whether other labs show the same pattern, or whether faster progress translates into a predictable schedule for future models.

Access, Patching and Further Evidence

OpenAI says GPT-6 Cyber will be available to approved Daybreak Red organisations, with access expected to reach all Red customers over the coming weeks. Its programme requires organisational and individual security checks, with API access restricted by default to approved IP addresses. Future information about model use, additional vulnerability disclosures and patch status may help clarify how the reported Linux work was handled.

Security teams should monitor advisories from software vendors and maintainers and apply relevant fixes promptly. That advice does not depend on every OpenAI figure being independently confirmed: the broader concern is that tools capable of finding and testing flaws may reduce the time available to respond. For now, the scale of that change—and how much of METR’s reported acceleration is specific to Anthropic—remains a matter for further evidence.

Key Questions

What is GPT-6 Cyber?

OpenAI describes it as a cyber-specialized version of GPT-6 Astra for authorized security work, including vulnerability discovery and testing. Access is limited to organisations approved for the Daybreak Red programme.

Did GPT-6 Cyber produce 948 Linux exploits?

OpenAI says a mid-October campaign powered by GPT-6 Cyber generated 948 working local privilege-escalation exploits. The supplied material does not show an independent verification of the full count.

What did METR report about Anthropic?

METR said Anthropic’s main measured progress rate was 1.9 times faster after October 2025 than before. It estimated AI assistance may explain about 80% of the acceleration, while warning that much of its most informative evidence was redacted.

Can the public use GPT-6 Cyber?

OpenAI has restricted access to approved organisations in Daybreak Red. Its announced controls include identity checks, managed-device and authentication requirements, and limits on API access.

What should computer users do now?

Install security updates promptly and follow advisories from software vendors and maintainers. The reports do not establish that every system is exposed to the vulnerabilities described.

Source: ThorstenMeyerAI.com