OpenAI is decelerating development of its Astra model following an internal security review. The company determined that the system had advanced to a critical level in agentic programming and cybersecurity capabilities, with potential to autonomously identify and execute attacks against traditionally well-protected systems.
In response, OpenAI is reinforcing security controls and has begun testing the model with government agencies and selected AI security organizations. The move reflects the company's caution as agentic AI systems—models capable of independent decision-making and action—continue to raise questions about safety and containment.
