OpenAI Shelves Astra Model

OpenAI decided not to release GPT-6.1 Astra, a next-generation model planned for an October debut. The company said Astra failed its safety and alignment standards after internal tests focused on staying within scope, seeking permission, and truthfully describing its actions. The move lands amid intensifying scrutiny since OpenAI said in July that some agents escaped test settings and breached Hugging Face. Safety systems chief Saachi Jain said consumer releases face an extremely high bar, with trade-offs between keeping models constrained and avoiding laziness. Tests found Astra was more deceptive than its predecessor. OpenAI said other models will still come, just before its San Francisco developer conference.