EU AI ActArt. 50Deadline 2 Dec 2026

The transparency deadline that didn't move: a deployer's checklist for 2 December 2026

By BigJay Jang · former IBM & Deloitte strategy consultant

Most AI Act dates feel far away. This one doesn't — and it applies to ordinary companies, not just AI vendors. If you deploy a chatbot or publish AI-generated content, Article 50 already applies to you, and the clock runs out on 2 December 2026.

What changed, and why the date is hard

Article 50 sets the EU AI Act's transparency obligations, and enforcement began on 2 August 2026, overseen by the European Commission's AI Office and national authorities. The important twist: the grace period for existing generative systems was shortened from six months to three. That moves the hard compliance date to 2 December 2026 — and unlike some high-risk deadlines that slipped into 2027–2028, this one held. As we put it internally: the deadline that didn't move is the one that matters.

Provider vs. deployer. Some Art. 50 duties fall on the provider (e.g. marking AI output as machine-readable at the source), others on the deployer — the company that puts the system in front of users. Most mid-market companies are deployers, and deployer duties are the ones people miss.

The deployer checklist

  • Chatbot & AI-interaction disclosure. If people interact with an AI system, they must be told they're dealing with AI — unless it's obvious from context. A clear notice at the start of the interaction is the safe default.
  • Deepfake labelling. If you publish image, audio or video content that has been artificially generated or manipulated to resemble real people, places or events, it must be disclosed as such.
  • Public-interest text. AI-generated text published to inform the public on matters of public interest must be disclosed as AI-generated, subject to the Act's conditions.
  • Emotion recognition & biometric categorisation. If you use these on people, they must be informed.
  • Synthetic-content marking. AI-generated or manipulated output should carry a machine-readable mark; providers carry the core duty here, but deployers should confirm the tools they use actually do it.

Why this bites harder than it looks

Two reasons. First, penalties are real: the AI Act's fine tiers reach up to €35M or 7% of global turnover for the most serious breaches, with lower tiers (€15M/3% and €7.5M/1%) for others (Art. 99). Second, the obligations attach to use, not to being a tech company — a marketing team publishing AI-generated visuals, or a support team running a chatbot, is squarely in scope.

How this connects to ISO 42001

Article 50 tells you what must be true; it doesn't give you a system to make sure it stays true as you add new tools. That's where an AI management system earns its keep: transparency becomes a control you own (mapped in ISO 42001's A.8), with evidence you can show. Meet the Act and build the certifiable core off the same work.

Check your transparency controls →

Get the monthly regulatory brief →

Sources: EU AI Act Articles 50 & 99; European Commission AI Office communications. Educational orientation, not legal advice; confirm your specific obligations with counsel. References current as of August 2026 — see the full sources & verification table.