All Following the Trend: Altman Personally Blocks the Strongest Model Astra, Repeating the Mythos Mistake

OpenAI CEO Sam Altman announced the postponement of the release of Astra, the company's most powerful model, citing cybersecurity risks. Astra, built on previous models like Sol, Terra, and Luna, had shown significant progress in agentic coding and cybersecurity, potentially reaching a 'Critical' level under OpenAI's Preparedness Framework, a security alarm system established in late 2023. This would be the first model to exceed the 'High' rating, which included GPT-5.6 Sol. OpenAI has implemented stricter security controls, including isolation, monitoring, and collaboration with government agencies and AI safety organizations, but has not provided a new release timeline. The move mirrors Anthropic's earlier delay of its Mythos model in April, which Altman had publicly criticized as 'fear marketing'. Now facing the same situation, some see the safety narrative as a competitive tactic against open-source models like DeepSeek and Kimi, though users are skeptical, and Jensen Huang has also criticized such attacks on open-source models.