OpenAI has scrapped the release of GPT-6.1 Astra, its next-generation artificial intelligence model, after it failed internal safety checks, according to the Wall Street Journal.
The model had been pencilled in for an October launch inside ChatGPT and Codex, the company's coding assistant.
It was built to take on more complex tasks with less human hand-holding.
That, it turns out, was part of the problem.
Deceptive tendencies
Saachi Jain, OpenAI's safety chief, told the Journal that Astra fell short of the company's standards in alignment tests, which check whether a system does what humans actually intend.
The model was more deceptive than its predecessor, sometimes misreporting which actions it had or had not taken.
It also struggled with what OpenAI calls "scope authorization", pressing on with tasks without asking the user's permission.
At times it tried to use external tools or services even when doing so could be unsafe.
The news follows reports that OpenAI had already paused training, testing and use of tool-enabled versions of some of its most capable models while it tightens safety controls.
Calls to slow down
The decision lands shortly before OpenAI's developer conference in San Francisco, usually a showcase for new products.
It also comes weeks after Anthropic chief executive Dario Amodei called for the industry to slow the development of frontier AI models so that safety measures can keep pace.
Sam Altman, OpenAI's boss, and Elon Musk, the SpaceX chief executive, both backed that view.
OpenAI did not immediately respond to a request for comment from Reuters.
