OpenAI Halts Development of Next-Generation AI Model Astra Due to Potential 'Critical-Level Cyber Capabilities'

OpenAI has paused certain internal development activities for its upcoming AI model, Astra, after internal evaluations indicated that the model could not be ruled out as possessing 'Critical-level cyber capabilities.' This decision was made under the company's Preparedness Framework, which outlines protocols for AI models acquiring dangerous capabilities in domains such as cyberattacks. The evaluation is ongoing, and this marks the first time the 'Critical' threshold has been reached, triggering development constraints.