GPT-6 Astra Scores 100% on ExploitBench as OpenAI Blocks PoC Exploit Requests
OpenAI unveils GPT-6 Astra, achieving 100% on exploit development tests.
Run OpenAI?
Get an email when a reviewed story names OpenAI, usually within the hour.
Free. Your list stays private and never appears in a subject line. One click stops it. How Vendor Watch worksPrivacy
Summary
OpenAI has launched GPT-6 Astra, an AI model that scored a perfect 100% on ExploitBench, a test for developing exploits from software vulnerabilities. While the model demonstrated advanced capabilities in identifying and exploiting zero-day vulnerabilities, OpenAI has implemented safeguards to prevent its misuse for creating proof-of-concept exploits, limiting its current use to secure code review and patching.
The lesson from this incident
Full lessonGPT-6 Astra's perfect score on ExploitBench demonstrates that AI models have reached a capability threshold where they can autonomously identify and exploit zero-day vulnerabilities at scale, dramatically lowering the barrier for sophisticated attacks. While OpenAI has implemented guardrails to block direct proof-of-concept exploit generation, these controls are only as strong as their enforcement and can potentially be bypassed through prompt manipulation or via less-responsible AI providers. This development compresses the already-narrow window between vulnerability disclosure and weaponization, making rapid patching and proactive vulnerability management more critical than ever. Organizations must recognize that adversaries now have access to AI-powered offensive tools that can outpace traditional human-speed defenses.
What helps against this
How we list vendors- AI application security
Prompt injection, data leaks through AI tools and unapproved AI use get caught.