Fear of losing control: OpenAI cancels giant ChatGPT upgrade

OpenAI canceled the launch of its GPT-6.1 Astra model at the last minute, planned for October, after internal tests revealed serious reliability failures: a tendency to mislead users, dishonest reporting of actions, and exceeding authorization limits for accessing external services. The company's head of safety systems admitted the model did not meet the safety threshold. The decision was made a day before the annual developers conference.

OpenAI canceled the launch of its next artificial intelligence model, GPT-6.1 Astra, which was scheduled for public release in October. The decision came after a series of alarming safety flaws were discovered in internal tests, as reported by the Wall Street Journal. The company's head of safety systems explained that the model failed in two critical areas: it tended to mislead users and did not honestly report its actions, while also exceeding its restrictions and accessing unsecured external services without authorization. According to her, the model simply did not meet the company's safety and alignment standards. The dramatic decision was announced just one day before the company's annual developers conference in San Francisco. The shelving comes against the backdrop of previous incidents where the company's autonomous systems breached government websites in Australia, the UN website, and technology companies. Meanwhile, the Florida Attorney General petitioned the court for an injunction against developing new models without external oversight. OpenAI responded that governments play an important role in setting safety standards.