Analysis
The European Commission's AI Office began enforcing the AI Act's transparency obligations on August 2, requiring chatbots and virtual assistants to disclose that users are interacting with AI rather than a human, deepfakes and other synthetic content to be labeled, and emotion-recognition or biometric-categorization systems to warn people they're being analyzed, according to the European Commission and Cooley. Article 50's only carve-out is when the AI nature of an interaction is already "obvious" -- a standard the Commission's guidelines interpret narrowly.
Penalties for noncompliance reach 15 million euros or 3% of worldwide annual turnover, whichever is higher -- a real number for any company with meaningful EU revenue. Separately, providers of general-purpose AI models whose training compute exceeds 10^25 FLOPs, the threshold at which the Act presumes systemic risk, must now perform standardized model evaluations, assess and mitigate systemic risks, and report serious incidents to the AI Office. There's a transitional grace period until December 2 for the specific marking-and-detection technical obligations, giving providers already on the market a few extra months to build the labeling infrastructure.
“Penalties for noncompliance reach 15 million euros or 3% of worldwide annual turnover, whichever is higher -- a real number for any company with meaningful EU revenue.”
One week in, the practical effect for most US startups selling into Europe is a compliance checklist, not a business-model change: chatbot products need a disclosure surfaced in the interface, and any synthetic-media features need detection-compatible watermarking before the December deadline. But the bigger unresolved question is enforcement in practice -- the Commission hasn't yet brought a public case under the new rules, and how aggressively national authorities police smaller, non-EU-headquartered startups selling into the bloc is still untested.