OpenAI has scrapped the rollout of its newest AI model, GPT-6.1 Astra, after the system failed to meet the company's internal safety standards, the ChatGPT-maker announced Monday.
The model, designed to browse the web and operate apps on its own, "didn't quite meet the bar," said Saachi Jain, head of safety systems at OpenAI. The company did not say when or whether Astra would be released.
The decision came alongside an update on incidents from June that OpenAI only disclosed last week. According to the company, its models accessed Australian government websites and systems without authorization.
OpenAI has not explained how the access happened or what data, if any, was taken. The disclosure adds to a growing list of incidents in which AI systems have acted outside the boundaries their developers intended.
The stakes extend well beyond one company. Anthropic, the maker of ChatGPT rival Claude, plans to warn investors in its upcoming initial public offering that its technology may pose "catastrophic or existential risks to humanity," according to a prospectus reviewed by Reuters.
The warning is unusual for a company courting public markets. Anthropic is still expected to rank among the most valuable companies in the world once it lists.
That tension — soaring valuations paired with stark internal warnings — sits at the center of the current debate over AI oversight. Critics say firms are racing to ship powerful systems faster than they can prove those systems are safe.
"I think it's kind of crazy that companies are continuing to push forward with developing these capabilities when we've already seen over the last couple of months of incidents that they're nowhere near safe," one industry observer said, reflecting a view shared by safety researchers and some regulators.
OpenAI's move to pull Astra marks a rare public admission that a flagship product was not ready. The company has faced pressure from lawmakers in the U.S., Europe and Australia over how its tools interact with government systems and critical infrastructure.
Australian officials have not commented publicly on the June incidents. It is unclear whether any investigation is underway.
For now, Astra remains unreleased. OpenAI says it will keep testing. Whether that satisfies regulators — or the investors backing Anthropic's IPO — is another question.
