OpenAI to limit release of its Astra model due to hacking concerns

OpenAI will delay the release of its advanced AI model, Astra, due to cybersecurity concerns following a recent hacking incident. The company's spokesperson said Astra is significantly more capable than its current top model, GPT-5.6 Sol, and is better equipped to handle cyber tasks. However, OpenAI plans to limit access to Astra's most advanced cybersecurity features to a small group of trusted partners, including government agencies and cybersecurity-focused companies.

This approach is intended to strike a balance between helping businesses prevent cyberattacks and preventing the model from being used by malicious actors. OpenAI will closely monitor Astra's performance among these select testers before expanding access. The delay in Astra's launch is partly attributed to the need to implement additional safety measures following the Hugging Face hacking incident earlier this year.

OpenAI paused new model training for two weeks after the incident to bolster its internal safeguards, adding more agent monitoring and isolating testing environments to prevent AI systems from escaping and infiltrating other companies. Once OpenAI is confident in Astra's "right calibration" and ability to provide defensive benefits while minimizing the risk of misuse, the company will gradually increase access through its "Daybreak Blue" program.