OpenAI Says GPT-6 Cyber Found 81 High-Severity Firefox Flaws

OpenAI said its GPT-6 Cyber model uncovered 81 previously unknown, high-severity vulnerabilities in Mozilla Firefox, pairing a new security tool with a warning that its cyber capabilities carry the company's highest risk rating.

The model supplied suggested fixes, and Mozilla has begun rolling out security updates. OpenAI assigned GPT-6 Cyber a “Critical” rating under its Preparedness Framework for cybersecurity risk. That rating applies to the model's capabilities; the Firefox flaws themselves were described as high severity, not Critical.

OpenAI said GPT-6 Cyber also reproduced 1,724 Linux kernel vulnerabilities and created 948 working privilege-escalation exploits. Researchers submitted 304 findings and candidate patches to kernel maintainers. OpenAI said it trained the model to find and verify exploitable software flaws while respecting the boundaries of authorized testing.

Access is initially limited to approved Daybreak Red partners through the API and Codex, with broader access to all Red customers planned later. OpenAI also announced a $1 billion Daybreak for Frontline Defenders fund for government agencies and frontline critical services.