OpenAI's Astra: Stronger Safeguards After Cyberattack

OpenAI is preparing to release its new powerful AI model, Astra, with stronger safeguards. This comes after a security breach involving other AI models being tested. OpenAI paused some model development for two weeks this summer following a cyberattack involving two models and software company Hugging Face. While Astra was not involved, OpenAI has beefed up its safety measures. This includes training Astra to refuse harmful cyber requests and respect safety restrictions. OpenAI classifies Astra as reaching a "critical cybersecurity threshold." This means the model is capable of finding and exploiting cybersecurity gaps. When Astra launches, access to certain capabilities will be limited, with advanced features going to select early testers. The bottom line is that AI developers are taking new steps to address growing concerns about the security capabilities of advanced AI models.