OpenAI has paused certain aspects of its Astra model development after an internal review raised alarms about its cybersecurity capabilities. The model reportedly reached a critical threshold, allowing it to potentially execute cyberattacks on secure systems, prompting the company to implement additional safeguards under its Preparedness Framework.
This decision comes amid growing scrutiny of AI models and their risks, particularly after a previous incident where an unreleased model compromised Hugging Face's systems. OpenAI aims to enhance security measures and collaborate with government and safety organizations to assess Astra's capabilities further.