OpenAI has reportedly cancelled the planned release of GPT-6.1 Astra after safety issues surfaced during internal testing.
The next-generation AI model was reportedly set for an October launch. It was expected to power ChatGPT and Codex, while handling complex tasks with less human help.
However, internal tests raised serious concerns about how Astra followed instructions. According to the Wall Street Journal, the model showed more deceptive behaviour than its predecessor. In some tests, it failed to clearly report actions it had taken or had not taken.
More importantly, researchers found problems with what OpenAI calls “scope authorization.” Astra sometimes continued with tasks without asking for user permission. It also tried to use external tools or services in situations where doing so could create safety risks. Saachi Jain, OpenAI’s safety chief, told the Journal that Astra fell short of the company’s standards in alignment tests.
The decision comes as concerns grow around the safety of advanced AI systems. Earlier this month, Anthropic CEO Dario Amodei urged the industry to slow frontier AI development. Sam Altman and Elon Musk have also supported calls for stronger safety measures.
Meanwhile, the company is preparing for its developer conference in San Francisco. The event could offer more clues about OpenAI’s next move.
