OPENAI has unveiled GPT-6 Astra, described as the “world’s most intelligent and aligned model,” with the company highlighting exceptional cybersecurity capabilities. In its public briefing, Astra earned a perfect 100% on ExploitBench, the benchmark that tests a model’s ability to turn known software vulnerabilities into working exploits. This marks a significant jump from GPT‑5.6 Sol, which scored 78.5% on the same metric.
OpenAI also noted Astra’s stronger arbitrary code‑execution rates during tests involving flaws disclosed in the prior three months, including two zero‑day vulnerabilities in unspecified software.
Despite Astra’s demonstrated prowess, OpenAI emphasised the dual‑use nature of such tools: they can aid defenders in vulnerability discovery and patching, but could also lower barriers for attackers. Consequently, the current Astra release is restricted to secure code review and patching activities, with safeguards in place that block prompts intended to generate proof‑of‑concept exploits.
OpenAI said broader access will come later under its Daybreak initiative, which aims to boost defender capabilities with frontier AI, including a pilot with MS‑ISAC to supply frontline public‑sector defenders with Daybreak access. Astra’s rollout is currently limited to a small set of organisations, with wider availability planned for ChatGPT Plus, Pro, Business and Enterprise users, and via the OpenAI API, Azure and AWS Bedrock.