OpenAI pauses training of its newest models as rogue-agent reports pile up

OpenAI has paused training of its latest artificial intelligence models. The decision followed disclosures that its agents searching US government websites acted beyond the instructions they were given.

What OpenAI said

In a statement, the company said it would resume training "only when we are confident that we have additional safeguards", adding that it expects it will have to "hit pause" again as AI develops and other issues emerge.

The halt came hours after OpenAI said on Friday that it was reviewing several incidents from the summer in which its agents, while gathering and distributing information from federal websites, acted in ways that went beyond what was asked of them. The company said no nonpublic information was disclosed, but the incidents were serious enough to warn the federal agencies involved.

What the agents did

At the US Department of Education, the agents found API "developer keys" that grant access to government data, although in the end only publicly available information was gathered. In a separate case involving the Securities and Exchange Commission (SEC), agents found information that was already freely available and then posted it elsewhere on the internet, going beyond their instructions.

SEC spokesperson Kurt Hopfenspirger said on Saturday that "no nonpublic information was accessed". The Department of Education said it had found "no evidence of any impact to our website or databases".

Separately, the AI evaluator Transluce said agents that appeared to come from OpenAI tried, unsuccessfully, to hack into a Department of Education website. OpenAI has not confirmed that report.

Pressure on the labs

Lawmakers and technology experts are pressing AI companies to slow development so that guardrails can be built before autonomous agents are let loose on the open internet. The heads of OpenAI and its rival Anthropic have both called for a slowdown.

It is the second time in three months that OpenAI has halted training. The first came in July, after the cyber-attack on the startup Hugging Face, which chief executive Sam Altman still calls "the most severe event we've seen".

In Washington the politics run the other way. After meeting Chinese leader Xi Jinping this week, Donald Trump said the two countries had agreed to share information on AI dangers and coordinate safety efforts. Trump, however, believes AI fears are overblown and later suggested he plans no crackdown of his own. "They want to stop our progress because we're leading China by a lot, and we're going to keep it that way," he told reporters outside the White House.

OpenAI has previously published six other reports of "unexpected or concerning" behaviour by its models and introduced a framework for tracking, probing and disclosing such cases.