OpenAI launches GPT-6 Cyber, a security model for finding and fixing vulnerabilities
On November 5 OpenAI introduced GPT-6 Cyber, a model built for security research. According to OpenAI, it scores 48.5% on the ExploitGym benchmark, up from 32.1% for the previous GPT-5.6 Cyber, and was used to build working exploits for 948 Linux kernel flaws and to find 81 new high-severity bugs in Firefox. Access starts with a small group of approved customers of Daybreak Red, OpenAI's program for authorized security testing.
Read the original on OpenAI — GPT-6 Cyber ↗
Why it matters
AI models are becoming strong enough to find serious flaws in software that experts have checked for years. That helps defenders patch faster, but the same skills could help attackers, which is why OpenAI now rates the model "Critical" for cybersecurity risk in its own Preparedness Framework and limits who can use it.
What changes
- For developers
- Teams that maintain open-source or security-critical code should expect more bug reports found by AI, often with proposed patches. Mozilla has already shipped fixes for many of the Firefox issues.
- For businesses
- Companies get a faster way to test their own systems, but only through approved access with IP allowlisting and extra monitoring. OpenAI also points to a $1 billion grant program that subsidizes access for public bodies and essential service providers.
- For users
- For everyday users the effect is indirect: browsers and operating systems may receive security updates more often as these tools find and fix flaws.
Sources
- OpenAI — GPT-6 Cyber ↗ openai.com Primary source
Summarized by AI from official sources on November 6, 2026. Spotted a mistake? Let us know. How it works