OpenAI launches GPT‑6 Cyber for scoped vulnerability research

OpenAI on 5 November 2026 announced GPT‑6 Cyber, a cybersecurity model trained to find and validate exploitable vulnerabilities while staying inside the scope researchers set. Access is initially limited to the API and Codex for selected Daybreak Red customers.

Benchmarks and real-world research

On ExploitGym, GPT‑6 Cyber scored 48.5%, up from 32.1% for GPT‑5.6 Cyber — about a 51% relative gain, or roughly 16 percentage points. In a scope-alignment evaluation, it avoided out-of-scope exploits in 99.93% of tests without its deployment prompt and 100% with that prompt. OpenAI classifies the model as Critical for cyber under its Preparedness Framework (High for GPT‑5.6 Cyber).

Company researchers say the model helped identify 81 novel high-severity Firefox vulnerabilities (Mozilla is releasing fixes) and, in related pipeline work covered separately, develop 948 Linux local privilege-escalation exploits.

Limits

These are OpenAI’s own claims and evaluations; independent reproduction of the full suite is not reported here. Access remains restricted. The company frames the release as defensive research for authorized security teams, not a public exploit toolkit.