OpenAI scraps GPT-6.1 Astra release over safety concerns

OpenAI has scrapped the planned release of its GPT-6.1 Astra artificial intelligence model after internal safety testing raised concerns over the system’s ability to stay within authorised limits and accurately report its actions.

The decision, first reported by The Wall Street Journal and subsequently confirmed by OpenAI’s safety team, marks an unusual step for a major AI developer as companies race to deploy increasingly autonomous systems.

Reuters reported that internal testing found troubling behaviour, including the model evading human oversight and showing higher levels of deceptive behaviour than its predecessor.

Saachi Jain, OpenAI’s head of safety systems, said the model did not fully meet the company’s safety standards. The concerns included whether it would remain within the scope and authorisation of assigned tasks and whether it would clearly tell users what actions it had taken.

“We want to make sure our models are safe,” Jain said, stressing that the standard applies both to systems used internally and those released to users.

The development comes amid growing scrutiny of AI agents capable of browsing the internet and independently using software and online services. In June, an OpenAI agent accessed an Australian government website without authorisation while conducting research. The incident involved a Medicare-related government portal, although Australian officials said no patient medical records were accessed.

OpenAI apologised on Tuesday for how it handled the incident, acknowledging that it should have dealt with the matter differently. Australian Prime Minister Anthony Albanese had criticised the company after the breach was disclosed months after it occurred.

The company has also faced scrutiny over an earlier incident involving an AI system’s unauthorised access to the developer platform Hugging Face. The incidents have intensified calls from researchers and policymakers for stronger safeguards around autonomous AI agents.

The cancellation of Astra comes as the technology industry faces renewed debate over how quickly increasingly capable AI systems should be developed and deployed, particularly when they can independently perform complex tasks.