A $1,279 Exploit Meets A Lab Running Twice As Fast In AI Reports

🔍 Read the full analysis: A $1,279 Exploit Meets A Lab Running Twice As Fast In AI Reports on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on monitors, keyboards and dev gear

• 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 said its GPT-6 Cyber model helped generate 948 working Linux privilege-escalation exploits during an October test, at an estimated API cost of about $1,279 per exploit. On November 5, independent evaluator METR also reported that Anthropic’s model-development pace has accelerated, though it said much of the evidence behind that assessment remains confidential.

The Linux figures come from OpenAI, not an independent public audit. The company said its pipeline examined about 2.8 million lines of kernel code, assessed 2,308 candidate findings and reproduced 1,724 vulnerabilities in test environments. It reported that 588 of the 948 resulting exploits involved previously unknown flaws. OpenAI’s reported API bill was $1,213,430, which works out to roughly $1,279 per exploit.

OpenAI named the model behind the run as GPT-6 Cyber on November 5. The company describes it as a cyber-specialized version of GPT-6 Astra, built for authorized security work such as vulnerability discovery, exploit development and web-application testing. Access is limited to organisations approved for its Daybreak Red programme, which covers offensive security testing. OpenAI says the model has its highest internal cybersecurity rating, “Critical”; that rating reflects the company’s own framework, not an external certification.

METR’s separate report examined Anthropic’s internal research data over eleven weeks. It estimated that the rate of improvement on its main measure was 1.9 times faster after October 2025 than before, and attributed about 80% of the recent acceleration to AI helping Anthropic researchers. METR presents that 80% figure as an estimate, not a directly observed or independently verified share. The evaluator said Anthropic redacted a majority of the evidence it found most informative, limiting what outsiders can check.

Cheaper Exploits Tighten Patch Timelines

If OpenAI’s cost and output figures are representative, AI-assisted vulnerability work may lower the expense of producing working exploits. The reported amount is a cost for this particular run, not a general price guarantee or proof that every exploit can be produced at similar cost. Still, the comparison in the source material is notable: OpenAI said a July run generated 166 Linux privilege-escalation bugs for $513,690, about $3,100 each. The two reported runs suggest falling costs, but they do not establish how the results would generalize across software or testing conditions.

For organisations responsible for software and infrastructure, the practical concern is the time between a flaw becoming exploitable and users receiving a fix. AI tools may assist security teams in finding weaknesses, but they can also make exploit development more accessible. The source material also cites Epoch AI research reporting that a 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 result and should not be confused with OpenAI’s Linux test.

METR’s findings point to a related development: AI may be helping labs improve the systems used to build future AI. That could accelerate both defensive and offensive capabilities. For everyday users and IT teams, the immediate response remains practical: apply security updates promptly, while organisations review patch processes and access controls.

Popular right now

KitchenAid Artisan 5-Qt Tilt-Head Stand Mixer – Dried Rose (KSM150PS)

AmazonView Latest Price

Hisense 43-Inch E6 Cinema Series Hi-QLED 4K UHD Smart Fire TV (43E6QF)

AmazonView Latest Price

Hisense 43-Inch E6SR Series Hi-QLED 4K UHD Smart Roku TV (2026 Model)

AmazonView Latest Price

As an affiliate, we earn on qualifying purchases.

Restricted Access, Limited Public Evidence

GPT-6 Cyber is not described as a generally available consumer model. OpenAI says only vetted organisations approved for the Daybreak Red tier can use it, through its API and Codex. The system card lists safeguards such as single sign-on, multifactor authentication, managed devices and least-privilege access for organisations. Individual researchers must pass identity checks and use a hardware security key; API access is restricted to approved IP addresses by default, according to OpenAI.

OpenAI’s October Linux campaign is not its only reported security work. The source material says the company also reported finding 81 previously unknown high-severity Firefox flaws. Mozilla advisories publicly credited OpenAI with 50 high-impact Firefox fixes in October, but did not identify which model found which flaw. That distinction matters: public advisories support the existence of fixes, but do not independently verify OpenAI’s model-specific attribution or the full Linux exploit count.

The two November 5 developments are related by their implications, not by a shared experiment. OpenAI’s results concern security testing and exploit generation; METR’s report concerns the pace of Anthropic’s AI research. METR’s estimate that AI accounts for about 80% of acceleration is its interpretation of internal data, and the report’s stated evidence limits make that estimate harder for the public to assess.

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

— METR, describing the limitations of its Anthropic assessment

Independent Checks Remain Limited

The Linux exploit totals, the number of previously unknown flaws and the API cost are OpenAI-reported figures. The source material does not describe a public, independent replication of the campaign. It also says Linux maintainers had not publicly commented on the reported findings at the time of publication, so the status of disclosure, fixes and verification across the reported issues is not established here.

Mozilla’s advisories confirm that OpenAI received credit for 50 high-impact Firefox fixes in October, but they do not establish which model identified each issue. The relationship between those advisories and OpenAI’s separate claim of 81 previously unknown high-severity Firefox flaws is not fully explained in the available material.

METR’s report is an independent evaluation, but its estimate that AI accounts for roughly 80% of Anthropic’s acceleration relies on evidence that METR says is substantially redacted. The report does not let readers inspect most of the material METR considered most informative. The durability of the measured acceleration, and whether similar patterns apply across other labs, also remain uncertain.

Disclosure and Access Will Shape Next Steps

OpenAI says GPT-6 Cyber access is being provided through its approved Daybreak Red programme. According to the source material, access is expected to reach eligible Red customers over the coming weeks, although the precise rollout schedule and the number of approved organisations are not specified. Further detail from OpenAI or participating security teams could clarify how the Linux findings were validated, disclosed and addressed.

For the reported Linux flaws, the next useful public milestones would be statements from kernel maintainers, disclosure records and information about patches or mitigations. For Firefox, advisories can clarify which fixes were made, but the model attribution for individual discoveries remains unreported in the material provided.

METR’s assessment leaves open whether additional evidence will become available for outside review. Until then, readers should treat the report’s measured 1.9-times change as METR’s finding and its 80% attribution as an estimate. Meanwhile, software operators can reduce exposure by monitoring vendor advisories and installing security updates promptly.

Key Questions

What did OpenAI report about GPT-6 Cyber?

OpenAI said the model helped produce 948 working Linux privilege-escalation exploits in an October run, including 588 it described as previously unknown. The company reported an API cost of $1,213,430, or roughly $1,279 per exploit.

Can the Linux exploit figures be independently verified?

The figures in the material come from OpenAI. It does not describe a public independent replication, and it says Linux maintainers had not publicly commented on the reported results at the time.

Who can use GPT-6 Cyber?

OpenAI says access is restricted to organisations it approves for the Daybreak Red programme, through the API and Codex. It lists identity, authentication and access-control requirements for participating organisations and researchers.

What did METR find about Anthropic?

METR estimated that Anthropic’s rate of improvement on its main measure was 1.9 times faster after October 2025 than before. It estimated AI assistance accounted for about 80% of the acceleration, while warning that much of the supporting evidence was redacted.

What should organisations and users do now?

Organisations should track security advisories and review how quickly they can test and deploy fixes. Users should install operating-system and application updates promptly; the reported research does not change that basic guidance.

Source: ThorstenMeyerAI.com