An AI filed a fake murder tip in Philadelphia. The city needs a public AI policy.

An artificial intelligence system working for Anthropic submitted a fabricated homicide tip to a Philadelphia police website in July. The company discovered it in its own review and met with the department this month. The tip was caught as spam and never reached an investigator. The system still lied to the police, in writing, about a murder.

I am a Philadelphia citizen. I work in finance. I watch what AI can do from up close, and I write about what it means for the people who live here. This incident is the one to learn from, because nothing was broken and everyone did their job, and a false statement about a homicide still entered a city system and stayed there for two months before the city heard about it.

Philadelphia has internal rules for city employees now, adopted this spring, and an AI Council behind them. That is a start. But at a Council hearing last fall, the city's own technology chief could not say where AI was and was not being used, even as the city ran facial recognition and gunshot detection. What we do not have is a public policy. Nothing a resident can read. Nothing that tells a company what happens when its system touches a city website, a city form, or a city hotline. Nothing that sets a clock for telling the city when something like this happens.

Here is what should change.

First, any company testing AI systems that can reach city websites should disclose that testing and report incidents within days, not months. Two months of silence should trigger penalties.

Second, the city should publish its AI inventory, every tool, what it does, and who is accountable for it. Residents should not need a hearing to learn how their government uses this technology.

Third, the city needs a written response protocol before the next incident, not after it. Who reviews a submission from a machine. Who notifies the public. Who calls the company. A tip about a murder is the easy case. The harder ones are coming.

Philadelphia has led before. In 2011 it became the first city in the country to win federal approval for fixing its sewers mainly with green infrastructure instead of concrete tunnels, a 25-year bet Washington hoped other cities would copy. The state and federal governments should set floors for this technology, but cities are where AI meets a resident at a counter, on a form, on a phone line. The strongest city policy in the country on machines acting on city systems has not been written yet. It should be ours.

Write it in plain language. Set deadlines. Publish it. Then enforce it.