GPT-6 Astra Is the First Model OpenAI Classifies as Critical for Cybersecurity

OpenAI has designated its GPT-6 Astra model as reaching the Critical threshold under the company’s internal cybersecurity Preparedness Framework. During expert testing, the model successfully identified previously unknown vulnerabilities in an operating system kernel and a web browser while creating functional exploits. This represents the first time a model has met the highest risk classification for cyber-offensive capabilities. The development signals a shift in AI safety monitoring as models demonstrate an advanced, automated ability to discover and weaponize software defects.