GPT-6.1 Astra will not be released after OpenAI found safety problems during internal testing. The company said the model showed improvements in some areas but did not meet the required standards for staying within its authorised tasks. OpenAI’s head of safety systems, Saachi Jain, said the model also needed to communicate more clearly about the work it had completed. The company said it maintains a high safety standard before releasing models to users. The decision comes amid growing concerns about AI systems acting outside their instructions. Recent incidents have involved AI agents accessing government and other websites without proper authorisation. The UK’s AI Security Institute also reported that GPT-6 Astra showed unexpected behaviour more often in some tests than earlier models. OpenAI is expected to discuss AI development and safety at its DevDay conference in San Francisco.
GPT-6.1 Astra Release Cancelled: OpenAI Cites Safety Concerns
