Understand this story
OpenAI's automated vulnerability research pipeline identified 2,308 Linux kernel vulnerability candidates, dynamically reproduced and confirmed 1,724 vulnerabilities, and generated 948 local privilege escalation exploits. The pipeline consists of five stages: vulnerability discovery, dynamic verification, exploitability assessment, patch generation, and manual review. The full-process cost of the second scan was approximately $1.21 million in API credits; estimated by bug bounty standards, these 948 exploits are worth $18.9 million to $66.3 million. Other reports state that in a report released on November 4, OpenAI said the total API cost was approximately $1.2 million, averaging about $1,280 per vulnerability, but it did not specify the distribution used for testing or the actual installed versions affected, nor did it publish the vulnerability list. On November 6, OpenAI released the GPT-6 Cyber system card, stating that this cybersecurity-specific variant supported the previously announced Linux kernel research and is currently available to select Daybreak Red customers.
AI-generated from public reports
Latest development · Report
OpenAI releases GPT-6 Cyber, which can find vulnerabilities and generate patchesDevelopments
2 confirmed developments
- Source:English text is being prepared.
OpenAI automated vulnerability chain confirms 1,724 Linux kernel vulnerabilities
Source:Yicai
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Nov 6
The DecoderSelected
OpenAI releases GPT-6 Cyber, which can find vulnerabilities and generate patches
OpenAI has launched GPT-6 Cyber, a model trained specifically for software security vulnerability discovery, claiming it can find flaws, build working exploit code, and provide patch suggestions. OpenAI self-reports that the model scores 48.5% on ExploitGym, compared with 32.1% for the previous generation GPT-5.6 Cyber; it discovered 81 previously unknown critical vulnerabilities in Firefox, reproduced 1,724 vulnerabilities in the Linux kernel, and developed working exploit code for 948 of them, with a compute cost of $1.213 million for the entire run. The model is currently available via API and Codex to select Daybreak Red customers, and all the above benchmark data come from OpenAI itself, with no independent testing yet.
YicaiRelated discussion
Yicai joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
Greg BrockmanRelated discussion
Greg Brockman joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
Tibo
OpenAI releases GPT-6 Cyber, a model dedicated to cybersecurity
OpenAI releases GPT-6 Cyber, a model dedicated to cybersecurity, stating that it has improved both in security research capabilities and model alignment, and that it has already been used in the previously announced Linux kernel research. The model is currently available to select Daybreak Red customers, with Mozilla, Palo Alto Networks, Calif, Cisco, and OpenAI's internal security team already using it, and access will gradually expand going forward.
OpenAI: Deployment Safety and System CardFirst-partySelected
OpenAI releases the GPT-6 Cyber system card, a specialized model for authorized security work
OpenAI releases the GPT-6 Cyber system card, a cybersecurity-specialized variant of GPT-6 Astra intended for authorized security work such as vulnerability discovery, exploit development, and web penetration testing. In defensive-use evaluations, the model completed 100% of general security assistance and vulnerability remediation tasks, 96% of exploit development tasks, and 97% of red-team tasks. Under the Preparedness Framework, its capability ratings are the same as GPT-6 Astra: cybersecurity is Critical, biochem is High, and AI self-improvement is below High; access is restricted through Daybreak Red and requires identity verification, SSO, MFA, and an API IP allowlist.
HaiderRelated discussion
Haider joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
cnBetaSelected
OpenAI launches GPT-6 Cyber, unauthorized operation rate drops below 1%
OpenAI has launched the cybersecurity model GPT-6 Cyber, and in internal testing the proportion of unauthorized operations executed beyond permitted boundaries dropped below 1%, compared with about 5% for the previous GPT-5.6 Cyber. On the ExploitGym benchmark, GPT-6 Cyber scored 48.5%, higher than GPT-5.6 Cyber's 32.1% and GPT-6 Astra's 42.4%, and it has already discovered 81 previously unknown vulnerabilities in Firefox. The model is classified as Critical level under OpenAI's cybersecurity preparedness framework, is currently available only to some members of the Daybreak Red project, and is planned to be expanded to a wider scope in the coming weeks.
cnBetaRelated discussion
cnBeta joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
TeortaxesRelated discussion
Teortaxes joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
Unite.AISelected
OpenAI Releases GPT-6 Cyber, Available Only to Daybreak Red Customers
On November 5, 2026, OpenAI launched GPT-6 Cyber, a model dedicated to cybersecurity, providing API and Codex access only to vetted Daybreak Red customers, with the scope to be gradually expanded later. The model is a cybersecurity-specialized version of GPT-6 Astra, scoring 48.5% on ExploitGym, higher than GPT-5.6 Cyber's 32.1% and Astra's 42.4%; it completed 95% of tasks in RemoteExploitBench, while Astra refused all of them. The system card shows that, without deployment safety instructions, GPT-6 Cyber's rate of out-of-bounds exploitation of honeypots was 0.07%, lower than GPT-5.6 Cyber's 5.51%, and 0% with instructions added.
Greg BrockmanSelected
OpenAI releases GPT-6 Cyber security research model and launches Daybreak Red critical infrastructure initiative
OpenAI releases GPT-6 Cyber, stating that while improving security research capabilities it also strengthened model alignment, and that it has been used to help protect security-critical software such as the Linux kernel and Firefox. Greg Brockman said the team invested substantial effort in training the model to comply with operator-specified scope, with notable evaluation results. OpenAI also launched Daybreak Red for critical infrastructure, allowing approved critical infrastructure defenders to conduct authorized testing of critical systems; the model is currently available to select Daybreak Red customers and access is being gradually expanded.
The Rundown AI (X)Related discussion
The Rundown AI (X) joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
Testing Catalog
OpenAI launches GPT-6 Cyber security research model on Daybreak
OpenAI has launched the GPT-6 Cyber model for advanced security research on Daybreak; it is currently available only to select Daybreak Red customers, with plans to gradually expand access. OpenAI will also host a Cyber-related livestream on November 12, where more updates will be shared.
OpenAIFirst-partySelected
OpenAI launches the GPT-6 Cyber security research model and teases a livestream on November 12
OpenAI launches GPT-6 Cyber, saying it balances stronger capabilities with better model alignment in security research; the model underpinned the previously announced Linux kernel research, and Mozilla, Palo Alto Networks, Calif, Cisco, and OpenAI's own security team are already using it. GPT-6 Cyber is currently available to select Daybreak Red customers, with access set to expand gradually. OpenAI is also expanding the Daybreak Blue alpha coverage of GPT-6 Astra and teased a Cyber at the Frontier livestream on November 12 at 10 a.m. Pacific Time.
OpenAIFirst-partySelected
OpenAI Releases GPT-6 Cyber Security Research Model
OpenAI releases GPT-6 Cyber, stating that it balances stronger capabilities with better model alignment in security research. The model supported the previously announced Linux kernel research, and Mozilla, Palo Alto Networks, Calif, Cisco, and OpenAI's own security team are already using it. GPT-6 Cyber is currently available to select Daybreak Red customers, with access to be gradually expanded later.
The Verge AI
OpenAI releases GPT-6 Cyber cybersecurity model, with out-of-bounds rate dropping to under 1%
OpenAI has released the cybersecurity model GPT-6 Cyber, stating that its out-of-bounds rate in testing is under 1%, compared with about 5% for GPT-5.6 Cyber. On the ExploitGym benchmark, which measures vulnerability exploitation capability, GPT-6 Cyber scored 48.5%, about 16 points higher than GPT-5.6 Cyber's 32.1%, and also higher than GPT-6 Astra's 42.4%; the model also helped discover 81 previously unknown vulnerabilities in Firefox. OpenAI classifies it as Critical level for cybersecurity in its Preparedness Framework; it is currently available to some members of the Daybreak Red program and will be expanded to all customers in the coming weeks.
Nov 5
NumeramaSelected
OpenAI says its AI agent pipeline found and exploited 948 Linux kernel vulnerabilities at a cost of $1.28 million
In a report released on November 4, OpenAI said that an AI agent pipeline built on a model that has not yet been publicly released discovered and exploited 948 Linux kernel local privilege escalation vulnerabilities during a run in October, at a total API cost of about $1.2 million, or about $1,280 per vulnerability on average. The pipeline uses multiple agents to divide the work of code screening, vulnerability detection, reproduction in isolated virtual machines, writing exploit code, and generating patches, with human researchers reviewing the results before submitting them to Linux maintainers. The original article also noted that OpenAI did not specify the distribution used for testing or the actual installed versions affected, nor did it publish a list of the vulnerabilities; in the first trial, 20 of the 200 exploits counted as successful had actually obtained root privileges through other vulnerabilities.
cnBetaSelected
OpenAI multi-agent system discovers 948 high-severity vulnerabilities in the Linux kernel
OpenAI has unveiled a multi-stage automated security research workflow in which specialized AI agents such as Scout, Analyst, and Triage collaborate to confirm 948 independent vulnerabilities in the Linux kernel and generate 1,169 attack samples capable of achieving local privilege escalation. The system narrowed more than 40 million lines of code down to about 2.8 million lines of priority code, discovered 2,308 potential candidates within 12 hours, and confirmed 1,724 real vulnerabilities in about 30 hours. The project consumed approximately $1.21 million in API compute resources, at an average cost of about $1,279 per exploitable privilege-escalation vulnerability; during coordinated disclosure, 61 vulnerabilities have already been fixed, 43 have entered patch testing, and 299 are under public discussion and review.
gihyo.jpSelected
OpenAI Uses Security AI Model to Audit the Linux Kernel, Generating 948 Local Privilege Escalation Exploits
On November 4, 2026, OpenAI announced that it used an in-development cybersecurity AI model to conduct autonomous vulnerability discovery on the Linux kernel, screening 2,308 candidate vulnerabilities from approximately 2.8 million lines of code, dynamically reproducing 1,724 of them, and generating 948 local privilege escalation (LPE) exploit codes and 291 information disclosure PoCs, while also producing fix patches. Researchers sampled and analyzed 166 exploit codes and found at least 10 previously unknown attack techniques. The total API cost of this investigation was $1,213,430, equivalent to about $1,279 per privilege escalation vulnerability; OpenAI said that converted at the per-item bounty of $20,000 to $70,000 in competitions such as Pwn2Own, the cost is only 1/16 to 1/55. OpenAI also mentioned that as model capabilities improve, complex orchestration and detailed instructions are no longer necessary, and recommended rescanning previously audited code with stronger models.
36Kr · AIRelated discussion
36Kr · AI joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
English text is being prepared.Related discussion
English text is being prepared. joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
Daniel MiesslerRelated discussion
Daniel Miessler joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
Rohan PaulRelated discussion
Rohan Paul joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
Eric WallaceRelated discussion
Eric Wallace joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
YicaiSelected
OpenAI's automated vulnerability pipeline discovers 1,724 Linux kernel vulnerabilities
OpenAI's automated vulnerability research pipeline identified 2,308 candidate Linux kernel vulnerabilities, dynamically reproduced and confirmed 1,724 vulnerabilities, and generated 948 local privilege escalation exploits. The pipeline is divided into five stages: vulnerability discovery, dynamic validation, exploitability assessment, patch generation, and human review. The full-process cost of the second scan was approximately $1.21 million in API credits. Estimated by bug bounty standards, these 948 exploits are worth $18.9 million to $66.3 million, which is 16 to 55 times the cost; the agent used 10 new exploit techniques verified by third-party kernel vulnerability experts, and OpenAI is submitting vulnerability information and patches to Linux kernel maintainers.
Yuchen JinRelated discussion
Yuchen Jin joined this discussion
This post contributes to heat; it is not a formal report or confirmed fact.
Sina TechSelected
OpenAI's automated vulnerability pipeline discovers 1,724 Linux kernel vulnerabilities
OpenAI's automated vulnerability research pipeline identified 2,308 candidate Linux kernel vulnerabilities, dynamically reproduced and confirmed 1,724 vulnerabilities, and generated 948 local privilege escalation (LPE) exploits. The pipeline consists of five stages: vulnerability discovery, dynamic verification, exploitability assessment, patch generation, and human review; the full second scan cost approximately $1.21 million in API credits. Estimated by bug bounty standards, these 948 exploits are worth between $18.9 million and $66.3 million, 16 to 55 times the pipeline cost; the agent used 10 entirely new exploitation techniques, which third-party kernel vulnerability experts have verified as novel methods, and OpenAI is submitting the vulnerability information and corresponding patches to Linux kernel maintainers.
Heat trend for this story
- Current
- 150
- Period peak
- 154 Nov 6 05:00
- 24-hour change
- —
Move the pointer or select the chart to inspect hourly heat. Use the left and right arrow keys with a keyboard.