OPENAI has paused the development of its new AI model, Astra, after internal testing revealed significant cybersecurity capabilities that could meet its Critical risk threshold for cybersecurity. This decision marks the first instance of an AI lab publicly halting model development due to cybersecurity concerns. The pause allows OpenAI to implement stricter security controls, including monitoring for high-risk actions, isolated testing environments, and enhanced encryption.
The Preparedness Framework guides this decision, designed to document and mitigate risks associated with advanced AI models. The company aims to ensure that such models are deployed safely and beneficially.