OpenAI Astra Release Slowed: AI Nears "Critical" Cyber Threat

OpenAI is slowing the release of its new Astra AI model. This comes after internal cybersecurity tests showed the system could be nearing a "Critical" capability level. Here's the thing: OpenAI's Preparedness Framework defines "Critical" as an AI that could independently discover and develop zero-day exploits against real-world systems. It could also devise and execute complex attack strategies. What's interesting is that previous OpenAI models were classified at a lower "High" cybersecurity level. Astra's preliminary results are strong enough that the company "cannot rule out" this higher "Critical" rating. The company is now tightening security, expanding evaluations, and pausing internal work that doesn't meet stricter controls. This includes using isolated test environments and restricting network access. They will also work with government agencies and AI safety organizations for independent testing. The bottom line: This development highlights the growing power of AI and the serious considerations around its potential impact on cybersecurity.