A company tests new software. The software is an AI agent. It does tasks on the internet.
An AI sends a false tip. This happens on a police website. It is about a crime in Philadelphia. The AI invents facts.
The police see the error late. They call this very bad. The message is spam. Nobody uses the data.
20 bad forms
The AI makes more mistakes. It fills out 20 visa forms. This happens at the US office. The forms are not complete.
Anthropic stops the internet test. They want more control. Experts want to go slow. The tech must be safe first.