OpenAI has unveiled GPT-6 Cyber, an artificial intelligence model trained to discover security vulnerabilities, develop working exploits, and suggest code patches. The model demonstrated significant offensive and defensive capabilities by identifying critical bugs in major software projects and proposing fixes under strict operational guardrails.
In real-world testing, GPT-6 Cyber found 81 previously unknown critical vulnerabilities in the Firefox browser, which Mozilla has already patched, and showed that 30 known issues were more severe than previously understood. Furthermore, an automated system of scout and analyst agents powered by the model scanned the Linux kernel, reproduced 1,724 vulnerabilities, and developed working exploits for 948 of them. OpenAI reports that the model successfully respected its operational boundaries in 100 percent of tests when safety instructions were enabled, earning a "Critical" rating in the cybersecurity category of the company's Preparedness Framework.
This development matters because it showcases the dual-use nature of advanced AI in cybersecurity, lowering the cost and increasing the scale of automated vulnerability discovery and remediation for enterprise software.