🔍 Read the full analysis: How To Understand The $1,279 Exploit And METR's Faster-Lab Report on ThorstenMeyerAI.com
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
TL;DR
OpenAI says an October Linux-kernel test produced 948 working privilege-escalation exploits at an average API cost of about $1,279 each; 588 were previously unknown, according to the company. Separately, METR reports that Anthropic’s measured AI capabilities have advanced about 1.9 times faster since October 2025, while warning that much supporting evidence is redacted. The figures point to faster vulnerability discovery and AI-assisted development, but remain subject to important verification limits.
OpenAI says a mid-October test using AI agents produced 948 working Linux privilege-escalation exploits, at an average API cost of about $1,279 per exploit. The company named GPT-6 Cyber as the model behind the run on November 5, the same day the independent evaluator METR reported that Anthropic’s measured model progress has accelerated; both developments raise questions about how quickly security flaws can be found and how much AI is speeding AI research.
OpenAI’s November 4 account says its pipeline examined about 2.8 million lines of Linux kernel code, identified 2,308 candidate findings and reproduced 1,724 vulnerabilities in test machines. It reported 948 working local privilege-escalation exploits after grouping findings by root cause. OpenAI says 588 were previously unknown. These are company-reported results; the source material does not include an independent audit of the full run or its accounting.
The total API bill was reported as $1,213,430, which works out to roughly $1,279 per working exploit. A comparison with an earlier OpenAI run illustrates the claimed cost change: in July, the company reported 166 Linux privilege-escalation bugs at a cost of $513,690, or about $3,100 each. The comparison uses OpenAI’s reported totals and does not establish the cost of finding or fixing every vulnerability across other systems.
OpenAI describes GPT-6 Cyber as a cybersecurity-focused version of GPT-6 Astra for authorized security work, including vulnerability research and exploit testing. Access is restricted to organizations approved for its Daybreak Red programme. The model’s “Critical” cybersecurity rating is OpenAI’s own assessment under its Preparedness Framework, not an independent certification.
GPT-6 Cyber · the Linux run
948 working exploits, about $1,279 each
OpenAI’s agents worked through the Linux kernel in mid-October. Every number on this board comes from OpenAI; our own arithmetic is marked ≈.
API cost per working Linux privilege-escalation exploit in OpenAI’s October run: $1,213,430 for 948 exploits. OpenAI’s own figures.
- Candidate findings analysed2,308
- Reproduced in test machines1,724
- Working exploits948
- …previously unknown588
About 2.8 million lines of kernel code scanned (≈7%). 948 = unique flaws after grouping 1,169 exploits by root cause.
Three months, one curve
OpenAI Linux runs, cost per privilege-escalation bug
- July 2026 · GPT-5.6 Sol166 bugs for $513,690≈ $3,094
- October 2026 · GPT-6 Cyber948 exploits for $1,213,430≈ $1,279
- Past hacking-competition payoutsper Linux privilege-escalation exploit$20k–70k
What is and is not known
As of 6 Nov 2026
- Reports go to kernel maintainers with candidate patches; a human checks each one first (OpenAI).
- Technical details, including at least 10 new techniques, are withheld until fixes ship.
- No public statement from Linux kernel maintainers found yet.
- Validated flaws with suggested patches are not the same as shipped fixes.
Sources: OpenAI, “A thousand fewer paths to root”, 4 Nov 2026, as quoted by The Hacker News and Decrypt; 588 previously unknown and the model as reported by Fortune, CSO Online and The Register, 5 Nov 2026. ≈ values: own arithmetic.
Powered by Thorsten Meyer AI
Lower Exploit Costs, Faster Patching Pressure
If OpenAI’s figures hold up, they suggest that producing working exploits could cost far less than traditional security research payouts. That does not mean every bug can be found for $1,279: the figure divides one reported API bill by one campaign’s exploit count, and does not include all research, validation, disclosure or remediation costs. Still, a lower technical cost could let more actors search for vulnerabilities at greater scale.
The defensive consequence is a tighter patching window. If vulnerabilities can be identified and turned into usable attacks quickly, organizations may have less time between a flaw becoming known and exploitation becoming feasible. Security teams need to track vendor advisories, test fixes and prioritize updates; users benefit from installing operating-system and application updates promptly. The report does not establish that the Linux findings have been exploited in the wild.
METR’s report adds a different concern: if AI tools are helping researchers improve AI systems, capability growth may compound. METR estimates that AI accounts for about 80% of the acceleration it observes at Anthropic, but says that estimate is uncertain and relies on evidence the public cannot fully inspect. The two reports concern separate organizations and activities; they do not show that OpenAI’s exploit work caused Anthropic’s faster progress.
Products Worth Considering
LG UltraGear 34GX90SA-W 34-inch WQHD Curved OLED Gaming Monitor
Samsung Galaxy Tab A11 (2026) – 8.7 90Hz Android Tablet, 64GB/4GB Wi‑Fi, Octa-core MediaTek Helio G99, Dual Dolby Speakers, 20W Fast Charger
DJI Osmo Mobile 8 3-Axis Phone Gimbal Stabilizer
As an affiliate, we earn on qualifying purchases.
What OpenAI and METR Measured
OpenAI’s Linux result concerns local privilege escalation: an attacker who already has ordinary access to a Linux machine could use a flaw to seek administrator-level privileges. It is distinct from a remote attack that begins over a network. The company says the tested code covered roughly 7% of the Linux kernel project, so the reported results should not be read as a complete scan of the kernel.
OpenAI has also described security work involving Firefox. The source material says Mozilla’s October public advisories credited OpenAI with 50 high-impact fixes, while OpenAI reported 81 previously unknown high-severity Firefox flaws. The advisories do not identify which model found each issue, so the figures cannot be directly reconciled from the available information. The source also cites Epoch AI’s late-October test in which the freely downloadable GLM 5.3 model reportedly turned published Firefox fixes into working attacks for about $1,000 per Firefox version. That is a separate evaluation, not part of OpenAI’s Linux campaign.
METR’s November 5 report examined Anthropic’s internal progress data over an 11-week review. Its main measure, as described in the source material, grew about 1.9 times faster after October 2025 than before. METR attributes roughly 80% of the acceleration to AI helping Anthropic researchers, but says Anthropic redacted a majority of the evidence METR considered most informative. Its finding is therefore an evaluator’s estimate based on limited public evidence, not a fully reproducible public dataset.
“The public has to rely on METR’s judgment much more than would be ideal.”
— METR, in its report on Anthropic
Independent Checks Remain Limited
The available source material does not show an independent audit confirming the 948-exploit total, the 588 previously unknown findings or the cost calculation. It also says no Linux maintainer had commented publicly at the time of the report. It is unclear how many findings were disclosed to maintainers, how they were triaged, how many have been fixed, or whether any were exploited outside testing.
Mozilla’s advisories confirm a set of credited fixes but do not map individual flaws to specific models. The relationship between OpenAI’s reported Firefox count and Mozilla’s advisory count is therefore unresolved. OpenAI’s access controls and cybersecurity rating are also company-described safeguards and judgments; the source material does not provide an external evaluation of their effectiveness.
METR says a majority of the evidence it found most informative was redacted. Its acceleration estimate and attribution to AI assistance should be treated as METR’s analysis, with the limitations the evaluator itself describes. The source does not provide enough public detail to independently reproduce the estimate or determine how it applies beyond Anthropic.
Disclosure, Patches and Further Evidence
The next practical milestones are clearer disclosure from OpenAI and Linux maintainers about the reported findings, including how many vulnerabilities were confirmed, patched or remain under review. Public advisories and maintainers’ responses would help distinguish the number of generated exploits from the number of distinct, actionable vulnerabilities and establish whether fixes are available.
OpenAI says GPT-6 Cyber is limited to approved Daybreak Red organizations, with access expected to expand to eligible customers over the following weeks. Its usage restrictions and monitoring will be relevant to track, but the source material does not say how compliance or misuse will be independently measured.
For Anthropic’s progress report, further public evidence or a fuller explanation of redactions could help outside researchers assess METR’s conclusions. Until then, the two reports provide signs of faster AI-assisted security research and model development, while leaving key details open. For organizations and individual users, the immediate step remains ordinary security hygiene: apply trusted updates promptly and follow vendor and maintainer guidance.
Key Questions
What does $1,279 per exploit mean?
It is the approximate average obtained by dividing OpenAI’s reported $1,213,430 API bill by 948 working exploits from its October Linux campaign. It is not a general price for finding or fixing a vulnerability, and the calculation does not capture every possible research or remediation cost.
Were all 948 Linux findings previously unknown?
No. OpenAI says 588 of the 948 were previously unknown. The source material does not provide an independent audit of that count or a complete public account of the findings’ disclosure and patch status.
Can anyone use GPT-6 Cyber?
No. OpenAI says access is restricted to organizations it approves for the Daybreak Red programme. Its system card describes identity, device-security and access-control requirements for participants.
What did METR conclude 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 accounts for about 80% of the acceleration, while warning that extensive redactions limit public scrutiny of the evidence.
Does the report show that Linux systems were attacked?
No. The source describes a test campaign that produced exploits in test machines. It does not report that the Linux findings were used in attacks in the wild.
Source: ThorstenMeyerAI.com
