Alan Turing asked whether you could tell a machine from a person. Europe has given up on that and made the machines introduce themselves instead.

The rules that landed earlier this month are simple enough. A chatbot has to say it is a chatbot. Deepfakes need labelling. An AI written article needs a tag unless a person actually read it before it went out. Banks will comply, because banks have compliance teams for exactly this. Newspapers will too. The person cloning a finance director's voice to push a payment through will not bother, and no law was ever going to change that. Which leaves things slightly backwards. The labels land on the legitimate content, while the material you actually need warning about turns up with nothing on it. Give it a year or two and an unlabelled message might start to read as the safer one. Does the labelling help, or does it just move where the trust sits?