Path to Astra: critical capabilities and frontier safeguards

OpenAI has developed a new model called Astra that meets the internal threshold for critical cybersecurity capabilities. Because of the model's high potential for misuse, the company will restrict access to these advanced features, providing them only to select partners. This decision represents a shift in deployment strategy, prioritizing the security of critical infrastructure and software defenses over a general public release. The move signals a new phase in how the company manages the risks associated with frontier artificial intelligence.