The EU AI Act Is Now Enforceable: Why Accessibility Became an AI Compliance Issue

On August 2, 2026, the EU AI Act's general application phase went live: market surveillance authorities in every member state now have formal supervisory powers, and one of the least-discussed consequences for web teams is that accessibility is officially wired into AI compliance. Article 16(l) of Regulation (EU) 2024/1689 requires providers of high-risk AI systems to meet the accessibility requirements of Directive 2019/882 (the European Accessibility Act) and Directive 2016/2102 (public-sector web accessibility); Article 4 (AI literacy) applies to any organization deploying AI in the EU — agencies included — and is now enforceable. If your site runs a chatbot, a copilot, generative recommendations, or synthetic media, you can no longer separate "accessibility" from "AI compliance." They are the same conversation.

What actually changed on August 2, 2026

The AI Act entered into force in August 2024 and applies in phases. The Article 5 prohibitions (unacceptable risk) and the Article 4 AI literacy duty have been in force since February 2, 2025. But August 2, 2026 is the big milestone: most remaining obligations became applicable, and — more importantly — the enforcement infrastructure went operational.

Three concrete changes, based on the detailed analysis Let's Copilot published on Article 4:

1. Market surveillance authorities are live across the EU: in Spain it's AESIA, in Germany BNetzA, and Ireland stood up its National AI Office. These bodies can already supervise, request information, and open proceedings.
2. Cross-border coordination is active: the AI Act's full enforcement architecture is running, so an investigation can start in one country and move across the Union.
3. Most obligations apply today: transparency (Article 50), data governance, technical documentation, and AI literacy. The exception is high-risk systems, which the Digital Omnibus — approved by the Council on June 29, 2026 — deferred to December 2, 2027 (Annex III standalone systems) and August 2, 2028 (Annex I regulated products).

The detail that worries us most as an agency: the Digital Omnibus softened Article 4 from "ensure a sufficient level of AI literacy" to "support its development." Legally, that moves it from an obligation of result to an obligation of effort. But it did not remove the duty, and it did not move the date. So: AI literacy is today the only AI Act obligation enforceable against organizations that don't operate high-risk systems — which includes nearly all of our clients.

Where accessibility enters the picture

Here's the point most AI Act coverage misses. Article 16(l) requires providers of high-risk AI systems to "ensure that the high-risk AI system complies with accessibility requirements in accordance with Directives (EU) 2016/2102 and (EU) 2019/882." The EAA and the Web Accessibility Directive are now, by reference, inside AI regulation.

And the EAA on its own is already in an accelerating enforcement phase. Level Access's analysis of the first year of enforcement confirms it:

• Germany: within weeks of the BFSG (Germany's EAA transposition) taking effect, e-commerce operators started receiving private warning letters from law firms. Germany allows private enforcement through competition law — you don't need to wait for a regulator.
• Netherlands: the ACM allowed voluntary self-reporting until October 15, 2025; since then, reporting is mandatory and the authority has sent information requests to e-commerce operators worldwide — including companies headquartered outside the EU that sell to Dutch consumers.
• Generally: civil society organizations have filed lawsuits, regulators are running market-surveillance inspections, and formal information requests are going out to global companies. And there's no cross-border cap: a site selling into five EU countries faces five potential investigations, one per national regulator.

The baseline is grim: the March 2026 WebAIM Million report found detectable accessibility failures on 95.9% of the world's top one million homepages — worse than the year before.

What this means for web teams building with AI

Our read at Mintec: the high-risk deferral creates a mental trap. Teams that conclude "this doesn't apply until 2027, moving on" forget that (a) Article 4 and transparency apply to everyone today, (b) the EAA applies today regardless of AI, and (c) enforcement will land first where it's measurable and where consumer complaints drive it: accessibility. An inaccessible chatbot is simultaneously an EAA problem and an AI Act problem. It's the perfect double violation — and it's exactly where regulators and law firms will look first.

The checklist we use at Mintec to audit AI-powered sites

In our accessibility engagements — the failure patterns we documented in our EAA audit article — we run an extended version of the process when the site has AI components. Four steps:

1. Inventory every AI touchpoint. Chatbots, copilots, recommendation engines, media generation, auto-summaries, lead scoring. Each has a different role (provider vs. deployer) and different dates. Synthetic media also carries its own transparency obligations — our AI Act watermarking analysis details them. No inventory, no compliance.
2. Map roles and market reach. Who supplies the model and who deploys it? Does the site sell to consumers in the EU? Territoriality is defined by the end user, not the company's HQ. The ACM already made that clear with its requests to non-European companies.
3. Audit accessibility on the AI surfaces. Chat widgets are among the worst offenders: trapped focus, no keyboard navigation, insufficient contrast, misused ARIA. The baseline is WCAG 2.1 AA (EN 301 549); if you're also building with well-formed native HTML you're ahead — we covered why native HTML is winning ground over ARIA in our earlier analysis. And if you're preparing for the WCAG 3.0 migration, our preparation guide for developers gives you the roadmap.
4. Document literacy and transparency. Team training register, AI usage disclosure on the site, human oversight mechanism. Article 4 is an obligation of effort, not result: what gets sanctioned is doing nothing.

Our take: cheap compliance exists, but it expires

The pattern is the same one we saw with the EAA in 2025: companies wait for the first warning letter. In Germany, an entire ecosystem of law firms already specializes in issuing them; in Spain, AESIA just activated its powers. Fixing a chat widget's accessibility today costs days of development. Fixing it after an official notice costs the same — plus legal fees, deadlines, and exposure.

The good news: the tools and standards exist. WCAG 2.1 AA, EN 301 549, native HTML semantics, and AI oversight practices any agency can ship in a sprint. What doesn't exist is more time: August 2 has passed, the EAA has been in force for over a year, and enforcement isn't waiting for the market to catch up.

The practical move: if your site — or your client's — runs any AI feature and sells (or intends to sell) in the EU, schedule an accessibility audit plus an AI inventory this quarter. This isn't an "accessibility project," it's product compliance. And as always with compliance: doing it right on time costs a fraction of doing it wrong later.