Could AI Labs Be Moving Faster? A Look At GPT-6 Cyber And METR

🔍 Read the full analysis: Could AI Labs Be Moving Faster? A Look At GPT-6 Cyber And METR on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get smart everyday buys 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 named GPT-6 Cyber on November 5 as the model behind a mid-October campaign that, by the company’s account, generated 948 working Linux privilege-escalation exploits, including 588 it said were previously unknown. The same day, independent evaluator METR reported that Anthropic’s model-development pace had accelerated, while warning that much of the evidence it reviewed was not public. Together, the developments point to faster AI-assisted security research and model development, but the underlying claims are not equally open to independent verification.

OpenAI describes GPT-6 Cyber as a cybersecurity-focused variant of GPT-6 Astra, intended for authorized security work such as vulnerability discovery, exploit development and web application testing. Access is restricted to organizations it approves for the Daybreak Red programme, which covers offensive security testing. The model is not being offered as a general public release. OpenAI’s system card says approved organizations must meet security requirements that include single sign-on, multi-factor authentication and managed devices; the company says it can monitor activity and revoke access.

OpenAI’s account of the Linux work says its pipeline scanned about 2.8 million lines of kernel code, analyzed 2,308 candidate findings and reproduced 1,724 vulnerabilities in test machines. It then generated 948 working local privilege-escalation exploits, with findings grouped by root cause. OpenAI says 588 were previously unknown. These are company-reported results; the source material says Linux maintainers had not publicly commented on the campaign, so independent confirmation of the count and status of the findings is not established here.

METR’s separate report says its main measure of Anthropic’s model-development progress grew 1.9 times faster after October 2025 than before. METR estimates that AI assistance may account for about 80% of the acceleration. That figure is an evaluator’s estimate, not a directly observed or publicly reproducible measurement: METR says Anthropic redacted a majority of the evidence it considered most informative. The report therefore offers a qualified assessment rather than a fully inspectable public record.

Cheaper Exploit Discovery, Faster Patching

The reported Linux results matter because they suggest that the cost of producing working exploits may be falling. OpenAI’s figures put the October campaign’s API bill at $1,213,430, or 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, about $3,100 each. These are company-reported costs and outputs, not a standardized independent comparison, but they point to a potential change in the economics of vulnerability research.

Lower costs can help defenders find and fix flaws, but the same capability can be misused. The source also cites Epoch AI’s late-October finding that a downloadable model, GLM 5.3, could turn published Firefox fixes into working attacks for about $1,000 per Firefox version. That separate result does not establish that GPT-6 Cyber’s Linux findings have been used maliciously. It does underline why restricted access and responsible disclosure matter, and why organizations may have less time between a vulnerability becoming known and an exploit being available.

For companies, the practical pressure is on vulnerability triage and patch deployment: discovering flaws faster has limited protective value if fixes are delayed. For individual users, the clearest action remains prompt installation of operating-system and browser updates. METR’s finding adds a second concern for the technology sector: if AI is helping researchers build more capable AI systems, development timelines may change faster than outside observers can evaluate.

Products Worth Considering

Canon SELPHY CP1500 Compact Photo Printer (White)

AmazonView Latest Price

HP OfficeJet 250 Wireless Mobile All-in-One Inkjet Printer

AmazonView Latest Price

Canon imagePROGRAF PRO-1100 17-Inch Professional Wireless Inkjet Photo Printer

AmazonView Latest Price

As an affiliate, we earn on qualifying purchases.

Two Reports, Different Evidence

OpenAI announced GPT-6 Cyber on November 5, after describing the Linux campaign on November 4. The model is the company’s fourth cyber-focused release this year, according to the supplied source, and its access rules are framed around approved security work. OpenAI assigns it a “Critical” cybersecurity rating under its own Preparedness Framework. That is the company’s classification, not an independent certification or outside test result.

There is some external material related to the security claims, but it does not verify every number. Mozilla’s public advisories credit OpenAI with 50 high-impact Firefox fixes in October, according to the source. The advisories do not identify which model found each issue, and they do not confirm the Linux exploit count. The evidence for the Linux campaign described here remains principally OpenAI’s own account.

METR is an independent evaluator, but independence does not remove the limitations created by restricted evidence. Its report compares Anthropic’s progress before and after October 2025 and attributes much of the change to AI assistance. METR also says readers must rely on its judgment more than it would prefer because Anthropic withheld much of the supporting material. The two developments are related by the broader question of AI accelerating technical work, but the OpenAI security campaign and METR’s Anthropic assessment are distinct findings.

“a majority of the evidence we found most informative”

— METR, on the limits of its evidence

What the Public Cannot Verify

The reported 948 Linux exploits and 588 previously unknown flaws come from OpenAI’s account. The supplied material does not include public confirmation from Linux maintainers, a complete independent audit of the results, or details showing which findings were accepted and fixed. It is also not clear from the material how the company’s root-cause grouping affects the exploit count.

For Firefox, Mozilla advisories credit OpenAI with 50 high-impact fixes in October, but do not say which model discovered each bug. That information cannot establish that GPT-6 Cyber found all, or any particular subset, of those flaws. METR’s estimate that AI accounts for about 80% of Anthropic’s acceleration is also uncertain because much of its most informative evidence was redacted. The basis and limits of that estimate should not be treated as fully settled from the public report alone.

Further questions include how consistently OpenAI can restrict use to authorized security work, how widely GPT-6 Cyber will be made available to approved organizations, and whether the reported costs and performance generalize beyond the specific campaigns described. No public account in the supplied material establishes misuse of this model.

Disclosure, Access and Follow-Up

OpenAI says GPT-6 Cyber access is limited to approved Daybreak Red organizations, with wider access for eligible Red customers expected over the coming weeks, according to reports cited in the source. The key near-term test will be whether additional technical details, security advisories or independent assessments emerge to clarify the Linux findings and their remediation status.

Readers should watch for comments from Linux maintainers and for advisories that connect discovered vulnerabilities to fixes without exposing users to unnecessary risk. For METR’s findings, further public evidence or later evaluations could clarify whether the reported acceleration persists and how much AI contributes. Until then, the results are best understood as a significant but qualified signal: security work and AI development may be speeding up, while public verification remains incomplete.

Key Questions

What is GPT-6 Cyber?

GPT-6 Cyber is OpenAI’s cybersecurity-focused variant of GPT-6 Astra, which the company says is designed for authorized tasks such as finding vulnerabilities and testing applications. OpenAI says access is restricted to organizations it approves through Daybreak Red.

Did OpenAI independently prove that 588 Linux flaws were new?

No independent confirmation is provided in the source material. OpenAI says 588 of its 948 working exploits were for previously unknown flaws, but the public evidence described here does not include a Linux maintainer’s confirmation or a full outside audit.

What did METR find about Anthropic?

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

Does the reporting show that GPT-6 Cyber was used in an attack?

No. The material describes OpenAI’s reported security research campaign and restricted access for approved organizations. It does not establish that GPT-6 Cyber was used in a real-world attack.

What should computer users do now?

Install operating-system and browser security updates promptly. The reports describe faster vulnerability discovery, but do not identify a specific newly disclosed Linux flaw that users should address directly.

Source: ThorstenMeyerAI.com