OpenAI to Restrict Astra Model After Rating It ‘Critical’ Cyber Risk

Internal testing found the new model capable of executing complex cyberattacks with minimal human input, prompting added security layers.