OpenAI Delays Next Major AI Model 'Astra' Over Critical Hacking Concerns

OpenAI has halted development of its upcoming AI model Astra due to concerns over its potential hacking capabilities. The company's latest internal evaluations revealed significant advancements in the model's agentic coding and cybersecurity, raising fears of critical cyber capabilities. Prior OpenAI models, including GPT-5.6 Sol, were categorized as "High," while Astra may have crossed into the "Critical" threshold.

This threshold is defined by the model's ability to identify and develop zero-day exploits in various critical systems without human intervention or to devise novel strategies for cyberattacks. To address these concerns, OpenAI is bolstering its safeguards and security controls, implementing isolated testing environments, sandboxed execution, and enhanced monitoring capabilities.

The company also plans to collaborate with government agencies and AI safety organizations during Astra's testing phase. OpenAI emphasizes its commitment to responsible deployment of frontier AI capabilities for the benefit of humanity. Notably, Astra solved 10 mathematical problems for a relatively low cost, showcasing the model's mathematical prowess.

This comes as AI models like Claude Mythos and Anthropic's Mythos uncover increasingly critical vulnerabilities in major tech companies like Apple, prompting Apple to tighten its bug bounty program due to the mounting volume of vulnerabilities being discovered.