Article 50 of the EU AI Act became enforceable on August 2, 2026, and with it the transparency regime that Brussels has been telegraphing for two years finally has teeth. Chatbots must disclose they’re AI. Deepfakes must be labelled. Generative outputs must carry machine-readable marks. Deployers of emotion recognition or biometric categorisation systems must tell the people being analysed.

Fines run up to €15 million or 3% of global annual turnover, whichever is higher. EU bodies themselves are capped at €750,000, a proportionality gesture that also happens to keep the Commission’s own institutions off the front page.

Enforcement is distributed across the European AI Office, national market surveillance authorities, and the European Data Protection Supervisor. That’s three overlapping jurisdictions, each with its own institutional priors, all now empowered to act against any provider whose outputs reach the EU market. The Commission adopted implementing guidelines on July 20, 2026, according to Cooley, and generative systems already deployed have a grace period running to December 2, 2026 to comply with marking obligations.

The softer route runs through the AI Office’s voluntary Code of Practice on Transparency of AI-Generated Content, which grants signatories a presumption of conformity and, per Cooley, a gentler enforcement posture. Several major providers have signed.

The timing is the tell. August 2 lands one day after the Trump White House’s 60-day AI framework deadline expired, per CNBC, leaving the two largest regulatory constituencies for frontier AI moving on parallel tracks that don’t touch. Brussels has chosen legibility through disclosure. Washington has chosen, for now, discretion. Every provider serving both markets now operates under two incompatible defaults.

Sources

Sources