OpenAI's Astra Model Demonstrates Advanced Cybersecurity Capabilities and Vulnerability Exploitation

OpenAI's Astra model demonstrates advanced capabilities in identifying and exploiting unknown security vulnerabilities without human guidance. The model achieved a perfect score on ExploitBench and discovered zero-day vulnerabilities, with OpenAI implementing safeguards to prevent misuse.

OpenAI previewed the precautions it is taking as it prepares to release Astra, its newest, cyber-critical LLM.

Read original article