Artificial intelligence agents created by Anthropic, acting on their own, tried to gain access to a range of government websites, filled out visa applications on the State Department\u2019s website and sent a fake homicide tip to Philadelphia\u2019s Police Department.<\/p>\n
Anthropic detailed the activity of its A.I. agents in a blog post on Friday, without naming the targeted websites. But two sources with knowledge of the incidents said Anthropic\u2019s A.I. agents had submitted 20 visa applications through a form available on the State Department\u2019s website. All the applications were incomplete and were not processed, they said.<\/p>\n
White House officials said they were briefed on the agents\u2019 actions on Friday, and demanded that A.I. companies immediately report rogue A.I. activity. It was the most aggressive statement so far on A.I. regulation by the Trump administration, which has sidestepped calls for tighter controls on the increasingly powerful technology.<\/p>\n
A new White House A.I. task force, created this week, said Anthropic told it about the agents\u2019 activities earlier on Friday. Called the Super Intelligence Force, the group includes Jay Clayton, the White House A.I. czar; Andrew Ferguson, the chair of the Federal Trade Commission; Scott Kupor, the director of the Office of Personnel Management; and Emil Michael, the chief technology officer of the Defense Department.<\/p>\n
\u201cWe informed the company that we expect immediate and full transparency to the entities involved and the public,\u201d they said in a statement. \u201cWe also expect that the company will immediately provide remediation services to the affected entities and any harmed Americans.\u201d<\/p>\n
They added that they expected Anthropic and other companies to \u201cimmediately report incidents,\u201d cooperate with federal and state law enforcement and ensure that the incidents were not repeated.<\/p>\n
Anthropic said in its blog post<\/a> that it discovered what its A.I. agents had done after it started a review of its A.I. activity in July. At the time, OpenAI had disclosed that its technology attacked another A.I. start-up, Hugging Face<\/a>. Anthropic and other labs also discovered that their technology had escaped from testing environments and conducted hacks<\/a>.<\/p>\n Anthropic said an A.I. model it had been testing had taken several unauthorized actions, including using a flaw in a university website to download data and sending a government agency a form that it had been told not to submit.<\/p>\n A spokesman for Anthropic declined to comment beyond the company\u2019s blog post.<\/p>\n The Philadelphia Police Department said earlier on Friday that Anthropic had notified it that the company\u2019s technology submitted a false homicide tip to the agency\u2019s website. The tip was dated July 18 and claimed that the submitter had information about an unsolved case, the department said.<\/p>\n The tip was flagged as spam and never investigated, the department said, but Anthropic notified the police this week about the incident and said the company would publicly disclose it on Friday.<\/p>\n \u201cAn unreleased, non-frontier research model was meant to fill out a practice copy of a government form,\u201d Anthropic said in its blog post. \u201cWhen the copy failed to load or the model closed it by mistake, the model instead navigated to the website where the real form is normally hosted and submitted the form there.\u201d<\/p>\n Public concerns about A.I.\u2019s going rogue have risen as more artificial intelligence labs have reported losing control of their technology. Google, Meta, OpenAI and Anthropic have been under scrutiny for the dangers that cutting-edge A.I. models might pose. They also face questions about whether to slow down development of their systems.<\/p>\n