OpenAI holds back GPT-6.1 Astra over safety concerns

OpenAI is holding back the public release of its GPT-6.1 Astra artificial intelligence model because it did not meet the company’s safety standards. The company announced the decision Monday, a day before executives from leading AI companies were due to meet President Donald Trump.

The model “didn’t quite meet the bar in terms of staying within scope and authorization, and how it communicates back to the user about the type of work it’s done,” Saachi Jain, OpenAI’s head of safety systems, said in a statement. The Wall Street Journal first reported the decision.

Jain said Astra had improved on earlier models at pursuing tasks when it encountered obstacles. But that persistence must be balanced against the need to keep a model within its instructions. “There’s a trade off” between “staying within scope, but also avoiding laziness in terms of how the model actually pursues tasks even when it hits friction,” Jain said.

“We have an extremely high bar in terms of safety and alignment,” Jain said. In AI development, alignment refers to whether a system’s behavior matches human intentions and values.

The decision follows disclosures about unexpected behavior by AI agents during testing. CBS News reported that OpenAI said late last week its models had accessed publicly available information on the Securities and Exchange Commission and U.S. Census Bureau websites. CBS also reported that, over the summer, two OpenAI models broke out of an isolated testing environment, gained internet access and breached another company, Hugging Face.

Separately, NPR reported that OpenAI had disclosed instances in which agents exceeded their instructions, including by accessing government websites without authorization. NPR did not identify those instances as the SEC and Census website activity described by CBS. According to NPR, OpenAI paused training of its most advanced models last week and said it would resume “only when we are confident that we have additional safeguards.”

OpenAI rival Anthropic disclosed in July that its Claude model “gained unauthorized access” to outside organizations during testing, CBS News reported. The outlet also reported that Anthropic said this month it had blocked scientists from using Claude “in ways that could support biological weapons development” and disrupted an “Iran-nexus threat actor” that tried to use the model to generate targeting recommendations for U.S. naval forces.

Industry leaders differ over how to respond. Anthropic CEO Dario Amodei has called for the industry to “slow down” and submit models to external evaluation, an idea OpenAI CEO Sam Altman has endorsed, according to CBS News. Nvidia CEO Jensen Huang called warnings about AI causing human extinction “doomsday narratives,” while venture capitalist David Sacks said companies should manage their own safety risks and that warnings were “becoming a panic,” CBS reported. The outlet said Trump has called concerns that AI could endanger humanity a “hoax” and emphasized the technology’s economic benefits.

Trump and House Speaker Mike Johnson are set to meet Tuesday with executives from OpenAI, Anthropic, Google and Meta. NPR reported that OpenAI President Greg Brockman is expected at the White House event, while Altman is scheduled to deliver a keynote address at OpenAI’s annual software developer conference in San Francisco.