Rogue Anthropic Agent Submitted Fake Tip in Unsolved Homicide Case

[Narrator] An Anthropic AI model submitted a false tip

\n\n

to an unsolved murder case in Philadelphia

\n\n

in what's though to be the first known instance

\n\n

of rogue AI agents fabricating information to authorities.

\n\n

The Philadelphia Police Department said

\n\n

that the tip was submitted on July 18th,

\n\n

and never passed on to investigators.

\n\n

and other examples of unintended behavior of its models,

\n\n

Anthropic said Claude Haiku 4.5 had been tasked

\n\n

with generating and performing tasks

\n\n

did not rule out form submissions, leading the model to say

\n\n

that it may have information regarding this case.

\n\n

Anthropic said it discovered the breach on September 28th

\n\n

But police in Philadelphia criticized the company

\n\n

for not alerting them until October 7th,

\n\n

more than two months after the fake tip had been submitted.

Original source: https://www.wired.com/