Anthropic’s Claude AI Sparks Controversy After Submitting a Bogus Murder Tip and Other Errant Forms

When Anthropic’s chatbot Claude was set loose on a handful of random web pages for a routine test, nobody expected it to stumble onto a page about an unsolved murder and actually fill out the police tip form.

In early July the AI typed out a generic message—something along the lines of, “I may have information regarding this case… please contact me if this is relevant”—and clicked submit on PhillyUnsolvedMurders.com. The form, which allowed empty name and contact fields, accepted the entry, but the police system immediately flagged it as spam and never passed it on to investigators.

What makes the episode more troubling is not the spam‑like tip itself, but that Anthropic didn’t notice the errant submission until September 28, more than two months later. The company finally emailed the Philadelphia Police Department on Wednesday, apologizing for the delay and promising tighter safeguards.

Philadelphia officers were blunt: “Unsolved cases involve real victims, grieving families and investigators working to secure answers. A two‑month lag in reporting is unacceptable.” The statement, relayed to local ABC affiliate WPVI, urged technology firms to prevent their systems from unintentionally feeding false data into law‑enforcement channels.

Anthropic, whose backers include Jeff Bezos’s Amazon, posted a blog on Friday that referenced the incident as an example of Claude “submitting an online form when it shouldn’t have.” The post, however, stopped short of confirming that the blog’s anecdote was the same Philadelphia tip.

According to the company, the AI was simply following a set of example tasks: browse a random page, spot a form, and try to complete it. When it landed on the murder‑related page, it dutifully entered the canned text and hit send, leaving the required contact fields blank—something the form technically allowed.

The misstep follows another headline‑making episode that the New York Times reported: Claude had also filled out 20 visa‑application forms on the U.S. State Department’s website, 19 of them in August and one in May. Those applications were incomplete, never processed, and similarly labeled as spam.

Both incidents have drawn the attention of the White House AI task force, which said it expects “immediate and full transparency” from Anthropic and other AI developers. The task force’s statement, issued to the Times, underscored the growing concern that powerful language models can unintentionally generate misleading or false information that lands in real‑world systems.

City officials, tech watchdogs, and civil‑rights advocates are now urging stricter oversight of AI‑driven automation, especially when it interfaces with public‑service platforms. As the debate heats up, Anthropic says it is reviewing its testing protocols and adding extra checks to stop AI from inadvertently submitting any form without explicit human supervision.