The EU just turned the AI Act from a compliance memo into a line item. As of August 2, 2026, the European Commission’s AI Office started enforcing transparency rules on AI models, with fines up to €15 million or 3% of worldwide annual turnover, whichever is bigger. For a startup doing $50M in revenue, that’s not a fine, that’s a board meeting.
The rules are specific: your chatbot has to say it’s a chatbot, your deepfakes need labels, and anything AI-generated has to carry machine-readable marks a detector can find. If you’re building on OpenAI, Anthropic, or Google’s models and shipping into the EU, this isn’t your vendor’s problem anymore. It’s yours, because the labeling obligation sits with whoever deploys the system, not just whoever trained it.
Here’s the part founders miss: the EU already gave you a compliance shortcut. Nearly 190 companies, including Anthropic, Google, Meta, and Microsoft, have signed the Code of Practice on Transparency of AI-generated Content, a voluntary framework that turns Article 50’s vague labeling language into a documented checklist regulators already recognize. Signing it doesn’t cost you engineering time you weren’t going to spend anyway, it just gets you a paper trail before an investigator asks for one.
That paper trail matters more than it sounds. The Commission opened a formal investigation into X over Grok in January after manipulated sexually explicit images, including content that may amount to child sexual abuse material, appeared on the platform, and that’s the model for what happens when you don’t have documentation ready.
Monday morning: check whether your product labels AI output automatically, and get someone to look at the Code of Practice before an EU regulator does it for you.
Nathan Zakhary