EU AI Act Article 50 Is Now in Force: What Changed on 2 August 2026
Article 50 is now in force. What changed under the EU AI Act on 2 August 2026, what's been delayed, and the grace period small teams keep confusing with their own deadline.
As of 2 August 2026, Article 50 of the EU AI Act is an enforceable transparency obligation. It is no longer a future requirement. If you publish AI-generated or AI-manipulated content professionally — text, video, audio, or social posts — the disclosure duty now applies to you. This article explains what that means in practice, and what it does not mean.
What actually changed on 2 August
The rule itself did not change. Article 50(4) and (5) read the same way since the EU published the AI Act in 2024. On 2 August, the obligation became applicable law, not a future date on a compliance roadmap. Deployers who publish AI-generated text on matters of public interest must disclose this clearly and distinguishably. The same duty applies to AI-generated or AI-manipulated deepfake-style image, audio, or video. Show the disclosure at the point a reader or viewer first encounters the content, not in a footer, in end credits, or behind a linked policy page.
If you already read the full disclosure guide and adjusted your publishing workflow, this date changes nothing for your operations. If you have not, note this line: content published from 2 August onward without a compliant disclosure creates a live compliance gap, not a future one.
The one grace period that matters
A genuine carve-out exists, and teams often miss it. The July 2026 “AI Omnibus” amendment (Regulation (EU) 2026/1744) gives providers of generative AI systems extra time. Providers with systems already on the market before 2 August 2026 have until 2 December 2026. In that time, they must bring content-marking mechanisms into compliance with Article 50(2). This machine-readable marking obligation sits on providers, not deployers.
Do not read this as a personal extension. It is a technical transition window for AI system providers to update their tooling. It does not delay your disclosure obligation as a deployer who publishes content. If you publish AI-assisted content from 2 August onward, your disclosure duty applies. This is true regardless of whether the tool you used finished rolling out its own machine-readable markers.
What didn’t happen on 2 August
Most people know another headline EU AI Act date: high-risk AI system obligations under Annex III (recruitment, credit scoring, education, and similar). That deadline did not arrive on 2 August. The same Omnibus amendment pushed it to 2 December 2027. If you delayed high-risk classification work because “the August deadline is coming,” you now have more time than the old date suggested. See the full deadline breakdown for details.
The rest of Article 50, briefly
The disclosure duty for public-facing content affects most small teams. Article 50 also covers more ground than deepfakes and text:
- §1 — chatbots and AI assistants: if you deploy an AI system that interacts directly with people (a chatbot, voice assistant, or automated support agent), you must make it clear they are talking to an AI. This does not apply when it is already obvious from context. Law-enforcement use is exempt, subject to safeguards.
- §3 — emotion recognition and biometric categorization: if you deploy either system, you must inform the people exposed to it. Any personal data processed must still comply with GDPR. This is exempt for lawful law-enforcement use.
- §4 exceptions: the deepfake and public-interest-text disclosure duties do not override evidently artistic, creative, satirical, or fictional work. Disclosure is still required, but must not interfere with displaying or enjoying the work. Content under a genuine editorial-responsibility review process (see the human-review test above) is also exempt.
- §6 — non-prejudice clause: none of this replaces or lowers the bar set by Chapter III for high-risk systems. It also does not override other transparency obligations elsewhere in EU or national law.
- §7 — codes of practice: the AI Office must encourage Union-level codes of practice to help everyone implement this consistently. One now exists: the Commission published the final Code of Practice on Transparency of AI-generated Content on 10 June 2026, and the Commission and AI Board confirmed it as an adequate voluntary tool for demonstrating Article 50 compliance. Adherence is voluntary; the underlying obligations are not. Non-signatories have to show their own measures are adequate. We signed it.
If you do not run a chatbot or biometric system, §1 and §3 do not affect your workflow. But learn the exceptions in §4 if you publish clearly satirical or fictional AI-assisted content and tend to over-disclose out of caution.
What to check today
- Audit anything you plan to publish or have scheduled — blog posts, social captions, video, podcast notes. Confirm the disclosure appears before the first line of substantive content, not after it.
- Check your editorial process honestly if you rely on the human-review exception. A light read-through before publishing does not qualify. The Commission’s transparency guidelines state this clearly at paragraph 135: “superficial, solely formal or procedural checks (e.g. spell-checking or grammatical correction), the mere existence of an editorial policy, automated review processes or cursory editorial approval without substantive engagement” cannot satisfy the condition. Human review means deliberate examination of the substance by someone with relevant expertise. Fact-checking is described as a minimum requirement.
- Watch what happens after sign-off. Paragraph 136 closes a gap in most workflows. If an AI system modifies, supplements, or reformulates the text after editorial review, the exception becomes void. Then you must treat the content as AI-generated. An AI-assisted headline rewrite or SEO pass applied after approval is enough to lose the exception.
- Do not confuse the provider grace period with your own deadline. If you are a deployer who publishes content, your date has arrived. The December extension belongs to the tool vendors, not to you.
This article is a deadline recap, not a new rulebook. The full guide has the placement templates, the EU’s official icon set, and the editorial-responsibility test in detail.
Related reading
See the EU AI Act checklist for small business to learn how Article 50 fits alongside the other five articles that apply to ordinary teams.
Want more guides like this? Browse all Perspectives or get in touch →
Found this useful? Share it or read more perspectives.