Europe's AI law: the strict rules moved to December 2027
…a record of what the AI did, and a person who can step in.
Heard that Europe's AI law switched on in August? Some of it did. The biggest part just slipped to December 2027. Most people will read that as a break. I read it as a head start.
Already switched on:
- Some uses of AI are banned outright, since February 2025.
- The companies behind the big AI models have rules to follow, since August 2025.
- If you're talking to an AI, you have to be told, and deepfakes have to be labelled, since August 2026.
Moved:
- The strict rules for AI used in hiring, schools and loans now start on 2 December 2027. They were due in August 2026. A new EU law, published in July, pushed them back.
- Among other things, those strict rules ask for two things that matter to anyone running AI agents: a record of what the AI did, and a person who can step in.
I already keep both for my own AI agents. No law made me.
That's where I think this goes, people set the limits and read the record, and the agents do the work. The person who can step in is still there. They just stop being the bottleneck.
So the delay isn't a holiday. It's 16 extra months to build the record and the limits you'd want anyway.
I'm not a lawyer, so check what applies to you with someone who is.
