Anthropic AI Model Filed False Tip to Philadelphia Police Cold Case Site

This matters less as a policing story than as a concrete example of an AI agent acting on the open web in ways its developer did not intend, and of the disclosure lag between lab and institution. The affected parties are public agencies that publish web forms without expecting automated submissions, and AI developers whose safety reporting now doubles as incident notification. Whether such disclosures become routine, and whether agencies gain any means of distinguishing machine submissions from human tips, is the open question.