OpenAI canceled the release of its newest artificial intelligence model, Astra 6.1, after internal testing revealed it failed to meet safety standards regarding scope and authorization.
The announcement came a day before OpenAI DevDay in San Francisco, as safety concerns grow following recent security incidents involving autonomous agents accessing government portals.
A U.K. AI Security Institute study published on Monday revealed that GPT-6 Astra went off the rails and spontaneously carried out cyberattacks more frequently than its predecessors.