OpenAI is deliberately slowing down the development of its upcoming AI model, Astra, due to concerns about its cybersecurity capabilities. The decision follows internal assessments that indicated the model’s potential to reach ‘Critical’ risk levels in cybersecurity and agentic coding.
Internal Evaluations Lead to Development Adjustment
Recent internal evaluations, supported by external expert opinions, prompted OpenAI to reconsider the pace of Astra’s development. The assessments highlighted risks that could not be ruled out under OpenAI’s Preparedness Framework, a safety protocol established in December 2023. This framework tracks AI advancements in fields like biology, chemistry, and cybersecurity.
Previous models, such as GPT-5.6-Sol, were rated at a ‘High’ risk level, but Astra’s capabilities suggest a possible shift to ‘Critical’. This change would mean the model might independently create zero-day exploits or develop new cyberattack strategies without human intervention.
Enhanced Security Measures and Testing
In light of these findings, OpenAI is enhancing its security measures and testing protocols to address Astra’s elevated risk profile. The company is implementing isolated testing environments, restricted access to tools and networks, and improved encryption measures. Additionally, OpenAI has paused any internal projects involving Astra that do not meet these stringent security standards.
A comprehensive monitoring system has been deployed for Astra, capable of analyzing the model’s cognitive processes and triggering immediate security responses to interrupt risky activities. This system covers both training and evaluation phases of the model’s development.
Collaboration and Future Plans
OpenAI is seeking collaboration with government entities and AI safety organizations to independently assess Astra’s capabilities. Recommendations for security controls will be shared with third-party partners involved in high-risk evaluations. This approach mirrors OpenAI’s previous actions in June 2025, when the company addressed similar high-risk thresholds for biological capabilities.
By ensuring that Astra’s advanced capabilities are developed responsibly, OpenAI aims to help defenders identify and fix vulnerabilities before they can be exploited by attackers. The company remains committed to working alongside governments, safety institutes, and civil society to ensure that advanced AI systems are deployed safely.
OpenAI continues to emphasize its dedication to transparency and collaboration, reinforcing its broader objective of advancing AI technologies while maintaining robust safety and security standards.
