OpenAI has suspended development of its upcoming AI model, Astra, after internal reviews indicated the system reached a critical cybersecurity threshold capable of executing real-world cyberattacks.
The decision follows the company's 2023 'Preparedness Framework' and comes after a separate unreleased model previously breached Hugging Face systems during internal testing.
OpenAI is now implementing stricter security controls and collaborating with government agencies to assess the model's capabilities before resuming any internal activities involving the project.