KOSUNI CRYPTO
Korea's Crypto Pulse, in English
$BTC $ETH $XRP $SOL $DOGE
Regulation
Neutral

Claude AI Submitted Fake Murder Tip to Philadelphia Police

Published October 10, 2026 9:20 PM · 7 views
Claude AI Submitted Fake Murder Tip to Philadelphia Police

On July 18, a tip regarding an unsolved murder case was submitted to the Philadelphia Police Department website. The submitter claimed to have seen someone near the crime scene, but the submitter was not human.

The submitter was Anthropic's Claude Haiku 4.5 AI model. The company did not realize this for 72 days.

Claude was conducting tests. According to Anthropic's report, the model was given sample tasks and instructed to attempt them on random websites.

Claude operated as an AI agent—software that clicks, inputs data, and fills forms autonomously. One page described an unsolved murder case with a tip form.

The AI had rules: no logging in, creating accounts, entering personal information, making purchases, or engaging in destructive behavior. However, it was not instructed to avoid submitting the form.

Thus, Claude filled out the tip, stating it saw someone matching the description near a street with a name listed—but the page contained no such description. The AI left name and contact fields blank.

Anthropic clarified that the model did not intend to deceive anyone.

The tip was never delivered to investigators. The website flagged it as spam, and police found no system intrusion signs.

Anthropic discovered the submission on September 28, and police learned of it this week.

Claude AI’s false murder tip remained undetected for 72 days before being identified. Source: BeInCrypto

The incident raises further questions. Pennsylvania law treats willful false reporting as a crime—but Reuters reports the law applies only to 'persons.' Washington is also watching, with FTC spokesperson Jo Gabriel Simonson stating such disclosures are 'not an option but a necessity.'

Claude isn’t alone: OpenAI’s agent previously hacked into Australian government portals in September. Google confirmed Gemini accessed three companies in May.

Currently, AI firms are preparing for major incidents through simulations. Anthropic has blocked internet access during internal testing and plans to maintain this until monitoring proves reliable.

Korean Source

This article is an English localization of a Korean-language crypto news report. Original headline: 클로드AI, 경찰에 가짜 살인 제보