Mark your calendar: 2 August 2026. That’s when Article 50 of the EU AI Act stops being theoretical and starts being enforceable.

If your organization touches generative AI in any way, chatbots, AI-written content, synthetic media, automated customer interactions, this is the deadline you can’t afford to miss. And the compliance teams that wait until July to figure out what it means will be scrambling.

Here’s what’s actually changing, and what to do about it before the clock runs out. 

Why This Is Happening Now

AI-generated content has gotten good. Too good, in some cases. Text, images, video, and audio that used to take a trained eye to fake can now fool almost anyone, which is exactly the problem regulators are trying to get ahead of.

Deepfakes. Synthetic media passed off as real. AI-generated misinformation spreads faster than anyone can fact-check it. Identity deception. A general erosion of trust in whether anything online is what it claims to be.

Article 50 exists to put a stop to the guessing game, giving people a clear way to know when they’re looking at AI-generated content or talking to a machine instead of a human.

What Actually Changes on August 2

Three obligations matter most, and they’re not optional for organizations in scope, even though the accompanying Code of Practice is voluntary.

AI-generated content has to be identifiable. If your organization is producing or manipulating content with AI, you need a way to mark it so people (and systems) can tell it’s synthetic. That applies across audio, image, video, and text.

Deepfakes need clear labels. Any content that’s been artificially generated or altered in a way that could mislead someone, especially anything that resembles a real person doing or saying something they didn’t, needs to be flagged plainly, not buried in fine print.

People have to know when they’re talking to a bot. Chatbots and other AI systems that interact directly with users must disclose that they’re AI, unless a specific legal exception applies. No more letting people assume they’re chatting with a human when they’re not.

The Code of Practice: Useful, Even Though It’s Optional

The European Commission published its Code of Practice on Transparency of AI-Generated Content in June 2026, and it’s worth reading even though nobody’s required to adopt it.

It splits guidance into two lanes:

• If you build generative AI systems, it covers how to mark and detect AI-generated or manipulated content.
• If you deploy generative AI systems, it covers how to label deepfakes and AI-generated or manipulated text in practice.

Adopting the Code isn’t mandatory, but it gives organizations a more predictable path to demonstrating compliance, which matters a lot when regulators start asking questions.

What Compliance Teams Should Be Doing Right Now

Waiting for the deadline is the wrong move. Start here instead:

1. Map where AI is actually being used across the business, not just where you think it’s being used.
2. Flag which outputs need disclosure or labeling. Not everything does; know the difference.
3. Update internal AI policies and governance documents so they reflect the new obligations, not last year’s assumptions.
4. Stress-test your controls for AI-generated media and deepfakes specifically.
5. Train the people actually producing content, marketing, customer service, comms, not just legal and compliance.
6. Keep watching for regulatory guidance. Implementation details are still taking shape.

The Bigger Shift

AI governance used to live almost entirely with the tech team. That’s over. Compliance, legal, risk, and executive leadership all have a stake in this now, and Article 50 is the clearest signal yet that transparency isn’t a nice-to-have anymore, it’s a baseline expectation.

Organizations that build transparency into how they use AI, rather than bolting it on after the fact, are the ones that will come out of August 2026 with less regulatory exposure and more trust with the people they serve.

The deadline isn’t the finish line. It’s the starting point for how AI governance gets done from here on out.