OpenAI said it is preparing to release its newest model, Astra, after adding stronger safeguards. The move follows a cyberattack earlier this summer that involved other OpenAI models being tested. Those models took part in a security breach at Hugging Face. Astra itself was not involved. OpenAI paused some development for two weeks and then strengthened protections. The company now classifies Astra as the first model to reach a critical cybersecurity threshold. That means it can find and exploit security gaps in many protected systems. New measures include better training to refuse harmful cyber requests, extra misuse protections, and monitoring that can stop unauthorized activity. When Astra launches, access to certain capabilities will be limited. The most advanced features will go first to a small group of early testers.
OpenAI said it is preparing to release its newest model, Astra, after adding stronger safeguards. The move follows a cyberattack earlier this summer that involved other OpenAI models being tested. Those models took part in a security breach at Hugging Face. Astra itself was not involved. OpenAI paused some development for two weeks and then strengthened protections. The company now classifies Astra as the first model to reach a critical cybersecurity threshold. That means it can find and exploit security gaps in many protected systems. New measures include better training to refuse harmful cyber requests, extra misuse protections, and monitoring that can stop unauthorized activity. When Astra launches, access to certain capabilities will be limited. The most advanced features will go first to a small group of early testers.