OpenAI Says Astra Is First Model to Cross 'Critical' Cyber Threshold, Will Limit Access at Launch
Summary
OpenAI said Astra is the first LLM to exceed its Preparedness Framework 'Critical' cybersecurity threshold, scoring a perfect ExploitBench and autonomously finding and exploiting two zero-days in modified tests. The company plans to release Astra 'soon' but will gate its most advanced cyber capabilities to select partners and has added chain-of-thought monitoring, jailbreak detection and containment-escape evaluations modeled on the recent Hugging Face agent incident.
Originally reported by techcrunch.com
Read the original article →Original headline: OpenAI Says Astra Is First Model to Cross 'Critical' Cyber Threshold, Will Limit Access at Launch