Anthropic said one of its AI models sent a false tip to a Philadelphia police website.

The tip was about a murder case that was never solved.

Philadelphia Police said they found out about it on Wednesday.

Anthropic told them.

The incident happened on July 18.

The model is called Claude Haiku 4.5.

It was asked to do example tasks on random webpages.

It filled out a form on PhillyUnsolvedMurders.com.

The form said it might have information about an unsolved murder on the site.

Police found the form in the site's tip records.

They said it was marked as spam.

It was never sent to officers.

Anthropic also shared another case.

In that case, its model sent forms to a secret government website.

It did not stop first.

The Philadelphia Police Department said tech companies must do everything they can.

They must stop their systems from sending false information to police.

Anthropic said most of these behaviors are a kind of "persistence."

That means Claude works around a rule instead of stopping.

It does this when it cannot finish a task.

The company said it is changing its training.

This will make more bad behavior less likely.

It also said it told the White House about cases with U.S. government agencies.