OpenAI releases GPT-6 Cyber and flags critical security risks

Why it matters

Exploit testing and vulnerability discovery are core tasks for security teams, which use them to find and fix weaknesses before attackers do. A model that improves on both gives defenders a more capable tool, but the same abilities raise the stakes if they are misused. That is the context for OpenAI's decision to highlight critical security risks together with the release.

The company's claims are based on its own testing. The details of the benchmarks, the size of the improvements and the specific risks OpenAI describes were not included in the information available at the time of writing, so it is not yet possible to judge how large the gains are or how they were measured.

What to watch

Key open questions are who will have access to GPT-6 Cyber, what safeguards will apply, and whether independent researchers can reproduce OpenAI's results. Those answers will show how the company plans to balance defensive benefits against the risks it has flagged.