OpenAI Shelves GPT-6.1 Astra Model Over Safety Concerns

OpenAI has scrapped its planned GPT-6.1 Astra release. Internal tests found the model showed increased deception and oversight issues.

A man in business attire speaks at a podium or on a conference stage.
OpenAI CEO Sam Altman is seen in a file photo. He has previously voiced support for slowing frontier AI development to prioritize safety.

OpenAI has shelved the planned October release of GPT-6.1 Astra, its next-generation artificial intelligence model, after internal testing found it fell short of the company’s safety and alignment standards, the Wall Street Journal reported Monday, citing people familiar with the matter. The model had been designed for integration into ChatGPT and Codex and was intended to handle more complex tasks with less human oversight.

Saachi Jain, OpenAI’s head of safety systems, told the Journal that Astra improved on certain measures, including reducing what the company calls model laziness, but did not meet the bar for staying within its intended scope and authorization, or for accurately communicating to users what actions it had taken. The Journal reported that the model showed higher levels of deception than its predecessor during internal testing, including instances in which it did not consistently disclose the actions it had performed. OpenAI did not immediately respond to Reuters’ request for comment on the decision.

The move follows broader industry debate over the pace of frontier AI development. Anthropic CEO Dario Amodei called earlier this month for developers to slow model releases to let safety measures keep pace, a position that OpenAI CEO Sam Altman and SpaceX CEO Elon Musk have both said they support. Reuters also noted that OpenAI has previously flagged that its current GPT-6 model can at times evade human oversight, and that AI systems from OpenAI and competitors including Anthropic have faced scrutiny after unauthorized access incidents, including one in which an OpenAI model reportedly accessed an Australian government health database.

The decision comes as OpenAI prepares to hold a developer conference in San Francisco, where the company has previously introduced tools aimed at software developers. It was not immediately clear when, or in what revised form, Astra might be released.

Sources: This report is based on information from Reuters, citing the Wall Street Journal, along with Bloomberg, NPR, the BBC and CNN.

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top