Learn about this first
OpenAI says its security researchers built an autonomous vulnerability discovery, exploitation, and patching pipeline powered by a cybersecurity model. In one run against the latest Linux kernel in mid-October 2026, the pipeline validated 1,724 vulnerabilities from 2,308 candidate findings, generated proof-of-concept exploits for 291 information-disclosure vulnerabilities, and produced working exploit code and candidate patches for 948 distinct local privilege-escalation vulnerabilities. OpenAI says the exploit code could run on stock Linux desktop configurations with standard security protections enabled, and that the agent also discovered and used at least 10 previously unknown Linux exploitation techniques.
OpenAI reported a total API cost of $1,213,430 for the run, or approximately $1,279 per local privilege-escalation vulnerability. Based on the Pwn2Own and TyphoonPWN bounty standards it cited, the hypothetical value of these vulnerabilities was $18.9 million to $66.3 million. OpenAI said it is working with Linux maintainers and Patch the Planet to submit vulnerability reports and patches, with human researchers reviewing the results before disclosure.
The numbers of vulnerabilities, patch effectiveness, novel exploitation techniques, and costs above all come from OpenAI’s claims and have not yet been independently verified. OpenAI subsequently released the cybersecurity model GPT-6 Cyber, saying its out-of-bounds rate had fallen from approximately 5% for GPT-5.6 Cyber to below 1%. It is currently available to some customers of the Daybreak Red project and will be expanded to all customers of the project in the coming weeks. OpenAI said the model supported the aforementioned Linux kernel research.
In addition, content reposted by Greg Brockman said that the model also discovered 111 renderer RCEs in Firefox, 81 of which were new vulnerabilities, along with exploitable exploits for 30 vulnerabilities known to developers but previously considered unexploitable.
AI-generated from reports · updated 11 days agoLatest developments
OpenAI releases cybersecurity model GPT-6 Cyber: 51% relative improvement on ExploitGym, 81 new Firefox vulnerabilities discovered
Popularity trend
Event progress6
OpenAI team uses a new model to scan OSS to protect internet infrastructure
Greg Brockman: OpenAI team used a cybersecurity model to scan open-source projects and discovered 111 Firefox renderer RCEsGPT-6 Cyber model released
OpenAI: Official update: OpenAI releases the GPT-6 Cyber cybersecurity modelOpenAI works with Linux maintainers to fix vulnerabilities
OpenAI: OpenAI works with Linux maintainers and Patch the Planet to discover and fix vulnerabilitiesOpenAI discloses novel Linux exploitation techniques discovered by an LLM
OpenAI: OpenAI says its agent discovered at least 10 novel Linux exploitation techniquesOpenAI introduces an automated vulnerability discovery and remediation pipeline
OpenAI: OpenAI’s automated vulnerability discovery and remediation processOpenAI publishes research on an autonomous Linux kernel vulnerability discovery and remediation pipeline
OpenAI: Official update: OpenAI uses a cyber model to automatically discover and patch 948 Linux kernel local privilege-escalation vulnerabilities