OpenAI Launches GPT-6 Cyber, a Critical-Rated AI Model That Finds and Fixes Software Vulnerabilities

OpenAI has released GPT-6 Cyber, a model purpose-built for offensive and defensive security work and classified "Critical" under the company's Preparedness Framework — the highest risk tier applied to any of its models so far.

What it does

The model discovers, validates, and proposes fixes for software vulnerabilities while staying inside researcher-defined permission boundaries.

Headline results

• Found 81 previously unknown, high-severity vulnerabilities in Firefox (now being patched by Mozilla).
• Built working local privilege-escalation exploits for 948 Linux kernel issues, hundreds previously unknown.
• Cost roughly 1.2 million dollars in API credits to run the discovery pipeline.

Benchmarks and safety

On ExploitGym it scored 48.5 percent, up about 51 percent from GPT-5.6 Cyber's 32.1 percent. It stayed within task scope 99.93 percent of the time without a safety prompt and 100 percent with one. Access is limited to vetted Daybreak Red customers.

Canonical blog: https://www.threerouter.com/blog/335