Anthropic Says Its AI Agents Have Filed Visa Applications, Also Reported A Murder

From the Washington Post, October 9:

Anthropic AI agents took ‘unintended’ actions on government sites
One AI system made a false tip to a police hotline.

SAN FRANCISCO — Anthropic, maker of the Claude chatbot, said that some of its AI agents had taken unintended actions on federal, state and local websites, including submitting a false tip about a murder to a Philadelphia police hotline.

The State Department said separately Friday that an Anthropic testing model had filed 19 visa applications in August and another in May using a form on the department’s website.

Anthropic said in a report published Friday that an internal review had discovered that its AI models had acted inappropriately in some instances during testing and while employees were using them.

In one incident, an AI agent exploited a design flaw in a state government website to freely access public data that usually required paying a fee. In another incident, an AI agent submitted a federal government form when it was instructed not to.

“We have briefed the White House on these cases and notified each agency involved,” Anthropic wrote in its report. The company said the incidents led it to turn off internet access for AI agents during all internal testing, pending a review of security measures.

The State Department said in a statement that the visa applications were incomplete and not processed....

....MUCH MORE

note to self: AI will be defeated by paperwork and bureaucracy. 

And at TechCrunch, October 9:

Anthropic can’t reliably control its AI agents. It’s cutting off its internal evals from the live internet instead