August 2 changed the rules
Not by banning AI, but by forcing it to introduce itself. Article 50 of the EU AI Act — its transparency obligations — became enforceable that day. Before, it was guidance. Now, regulators can actually fine you.
Picture a customer-service chatbot you built, selling to EU users. Historically you could sprinkle "I'm AI" wherever you felt like it. Skip it now and a penalty can land — and it isn't small.
What it actually demands
Three duties bite from day one. Any AI system that talks to a person must make clear the user is dealing with a machine. AI-generated or altered images, audio, and video need a machine-readable mark so downstream tools can detect them. Deepfakes must be labeled as such.
The penalty is plain: breach the transparency duties and you face up to 15 million euros or 3% of worldwide annual turnover, whichever is higher. Enforcement sits with each member state's market-surveillance authority, not a remote Brussels desk.
The trap almost everyone misses
Systems already on the EU market before August 2 got a grace period on content marking, running to December 2, 2026. But the chatbot-disclosure duty and the deepfake-labeling duty have no grace period — they applied immediately. Plenty of teams assumed the whole article was delayed and got caught on exactly the clauses that weren't.
The majors are already scrambling
Anthropic published an explainer on August 14 detailing the SynthID-Text watermark it will embed in future Claude outputs — squarely built to satisfy this rule. By late July, roughly 190 organizations had signed the official Code of Practice on transparency, a documented safe harbor. Signing isn't mandatory, but it makes proving compliance far easier.
Don't misread it as a high-risk rule
Article 50 is risk-agnostic. It keys off what a system does — interacts with people, generates synthetic content, recognizes emotion — not whether it's classified high-risk. A minimal-risk marketing bot is in scope; a high-risk resume screener that never talks to a candidate may not be.
Why this matters to you
If you ship an AI tool or site to overseas users, even as a one-person team, EU reach pulls you into this. Add an "AI-powered" notice in the UI now and mark AI-generated content. Don't wait for the first enforcement case to retrofit — by then the fine is already written.
Related reading: Anthropic, Claude