OpenAI says a prototype security pipeline, powered by an upcoming OpenAI cyber model and run in a controlled research environment, “identified and analyzed 2,308 candidate Linux kernel findings, dynamically reproduced 1,724 vulnerabilities”. The company published the research on 4 November 2026 and put the end-to-end cost of the October scan at $1,213,430 in API credits.
OpenAI counts 948 unique vulnerabilities that could let an unprivileged local user gain root access. By our arithmetic, that is about $1,280 per unique vulnerability, against about $3,094 in a July scan with GPT-5.6 Sol ($513,690 for 166), or about 59% lower. OpenAI calls its stage costs “more an upper bound”.
Business Pill · Verification Cost Explained
An evergreen Business Engineer explainer on why output that is cheap to make can be costly to check. It teaches the general idea only and does not discuss OpenAI’s post.
The key insight: As we read it, OpenAI’s post is a unit-cost disclosure as much as a security one: by our arithmetic the cost per unique root-access vulnerability fell from about $3,094 in July to about $1,280 in October, while the scan’s total spend rose from $513,690 to $1,213,430.
OpenAI describes a five-stage pipeline: vulnerability discovery, dynamic validation, exploitability assessment, patch generation and human review. Agents do the first four stages; “Security researchers review the exploit, patch, and proposed severity before reporting to the maintainers.”
The work extends June’s Patch the Planet results, when OpenAI said GPT-5.5 Cyber had found Linux kernel vulnerabilities and “developed 24 local privilege escalation exploits to validate and report them”. The upcoming model behind the October scan is not named in the post.
OpenAI frames the bottleneck as triage, not discovery. It writes that the challenge “was not discovery, but determining which findings could be reproduced, what security boundary they crossed, and which warranted the earliest maintainer attention”.
API credits by pipeline stage in OpenAI’s October 2026 Linux kernel scan, as stated by OpenAI on 4 November 2026; OpenAI calls the costs upper bounds. The 64% share and the stage sum are our arithmetic.
The Cost of Each Stage
OpenAI reports API costs for each stage. Scout agents picked 5,034 audit targets, about 7% of the kernel code base, for $3,426. Analyst agents produced the 2,308 candidate findings within 12 hours for $316,549. Validation reproduced 1,724 of them, about 74.7%, within 30 hours for $73,330.
An initial screening agent cost $44,194. The exploitation stage cost $775,929: $84,929 for information leak candidates and $691,000 for local privilege escalation candidates. By our arithmetic, the five stage figures add up to within $2 of the $1,213,430 total, and the exploitation stage is about 64% of it.
OpenAI adds that “the costs of attempting to validate a true positive and false positive were similar”, so removing false positives earlier makes the process cheaper. It also says failed exploitation attempts cost far more than successful ones, and that a give-up condition “would have reduced the cost of ExploitDev significantly”.
July vs October
OpenAI compares two scans of the latest kernel master branch. In late July 2026, “GPT-5.6 Sol found and exploited 166 unique LPE vulnerabilities for an end-to-end $513,690 cost in API credits”. In mid-October, the newer model “found and exploited 948 unique LPE vulnerabilities for an end-to-end $1,213,430 cost in API credits”.
By our arithmetic, October found about 5.7 times as many unique vulnerabilities for about 2.4 times the spend. OpenAI’s takeaway: “Rescan previously audited code as model capabilities improve.”
How Disclosure and Patching Work
OpenAI says it used the evidence to prioritize reports and developed candidate patches for maintainer review. “Following guidance from the Linux kernel community, we began with critical vulnerabilities and are working with Patch the Planet partners on further reports and patches.”
It reports that when the scan and manual analysis finished after 10 days, “61 LPEs had already been patched upstream, 43 had staging patches, and 299 were being reported and discussed publicly”. OpenAI says it has “already sent out hundreds of reports and patches for critical vulnerabilities”.
OpenAI says it withheld the full technical details of exploitation methods it describes as new “For safety reasons”. It also reports that 20 of 200 early exploit results reached root through a different vulnerability than the one assigned, and that it added an independent check after human review.
The Bug Bounty Comparison
OpenAI compares its cost with historical contest payouts of $20,000 and $70,000 for Linux kernel privilege escalation exploits. Applied to 948 exploits, it calls the result “a hypothetical valuation of $18.9M—$66.3M, approximately 16x to 55x our pipeline’s cost”.
A separate run against Google’s kernelCTF program cost roughly $32,936 and produced 21 privilege escalation exploits, which OpenAI says would mean about a $1.5M award if accepted. Its takeaway: the pipeline “was net profitable at the time of this research”, and “Defenders should proactively analyze their software so that it becomes cost prohibitive for attackers to exploit.”
As we read the cost table, the money sits after discovery: by our arithmetic about 64% of the $1,213,430 went on the exploitation stage, and OpenAI says the costs of validating true and false positives were similar.
As we read it, OpenAI’s own caveats limit how far the dollar figures travel: it calls them upper bounds, says it focused on methodology rather than cost, and expects harness engineering to cut the cost at each stage.
OpenAI, ‘A thousand fewer paths to root’, 4 November 2026
“We want to steadily reduce the opportunities available to attackers and raise the cost of finding and exploiting vulnerabilities, until the expected payoff no longer justifies the attack.”
Three Implications
FOR SECURITY BUDGETS OpenAI puts the October Linux kernel scan at $1,213,430 in API credits and calls its stage costs upper bounds; by our arithmetic that is about $1,280 per unique root-access vulnerability.
FOR OPEN-SOURCE MAINTAINERS OpenAI says that after 10 days 61 LPEs had been patched upstream, 43 had staging patches and 299 were being reported and discussed publicly, with critical vulnerabilities reported first.
FOR SOFTWARE BUYERS OpenAI writes that the rapid rate of disclosure and patching emphasizes the criticality for users and companies to keep software up to date.
The framework puts it this way: “The new job of finance is to manage the ratio between the two — work produced per token burned.”
As we read it, OpenAI’s two scans are that ratio in practice: by our arithmetic the October run found about 5.7 times as many unique vulnerabilities as July’s for about 2.4 times the API spend. The framework does not discuss OpenAI’s post or Linux security.
The figures are OpenAI’s own and come from a controlled research environment; we have not seen the underlying data. OpenAI says it focused on “methodology and measuring model capability in this research, not cost”, and that the costs are upper bounds. The post does not give a cost for patch generation, a price for the upcoming model, or a release date for it. The bounty figures are OpenAI’s hypothetical comparison, not payouts.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growthstory. This week it started being priced as a credit story.
OpenAI says its pipeline reproduced 1,724 of 2,308 candidate Linux kernel findings and found 948 unique root-access vulnerabilities for $1,213,430 in API credits. By our arithmetic, that is about $1,280 per unique vulnerability, about 59% less than its July scan.
A note on sourcing. We read OpenAI’s research post in full on openai.com (published 4 November 2026); every figure and quotation is from that post. The cost per vulnerability, the July-October comparison, the stage sum and the 64% share are our arithmetic. We have not seen OpenAI’s underlying data, and in line with our security policy we leave out technical and exploitation detail.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
