The Brief · July 31, 2026

Europe delayed the hard part of its AI law. The part that touches your chatbot arrives Sunday.

The EU pushed its toughest AI obligations out to 2027. But the one rule that lands on almost every business — you must disclose when a customer is dealing with AI — kept its August 2 date, reaches anyone serving European users, and is backed by fines that scale with your revenue.

A monumental glass-and-steel government institution facade at night, tall columns lit from within with cool teal light, an empty rain-slicked plaza mirroring the building

A law is a building most businesses never enter — until a deadline lights it up. On Sunday, the transparency floor of Europe's AI Act switches on, and its doors open onto every company whose AI touches a European customer.

What happened

On Sunday, August 2, a section of the EU AI Act called Article 50 becomes enforceable. It is the law's transparency layer, and it lands on ordinary businesses — not just the AI labs.

In plain terms: if a customer can talk to your AI, or read, see, or hear something your AI produced, you now have to say so.

The rule applies to any company whose AI reaches a person in the EU — including US businesses serving European customers — and breaches carry fines of up to €15 million or 3% of worldwide annual revenue, whichever is larger.

The AI Act has been on the books since 2024; August 2 is simply the day this particular obligation switches on. It is not a proposal or a consultation. It is a live compliance date.

The detail almost everyone will miss

Earlier this year, Europe agreed an "Omnibus" package that pushed its most demanding obligations far into the future. The rules for "high-risk" systems — hiring, credit scoring, critical infrastructure — moved to December 2027; the versions baked into regulated products slipped to 2028.

The coverage that followed was blunt, and for most readers it was the whole story: Europe delayed the AI Act.

But the one obligation that reaches almost every business — the duty to disclose AI — was left on its original August 2, 2026 date.

That is the trap. The delayed rules govern a narrow set of high-stakes systems most companies will never build. Article 50 governs the everyday stuff nearly everyone already runs: a support chatbot, a synthetic voice line, AI-written marketing copy, a generated product image.

The expensive, headline-grabbing part of the law got two more years. The cheap, universal part is the one arriving now.

A long empty institutional corridor at night receding into darkness, doors along both sides, cool teal and steel-blue light, a faintly reflective floor

Inside the rulebook, the demanding rooms — high-risk hiring, credit, infrastructure — were locked for two more years. The corridor everyone actually walks, the duty to say "this is AI," stayed open on schedule.

Why this matters if you run a business

Article 50 covers four concrete situations, and most operators sit inside at least one. A chatbot or voice agent that speaks with people must tell them they're dealing with a machine, unless it's already obvious. AI-generated images, audio, video, and text must be marked so they're detectable as synthetic. Emotion-recognition and biometric tools must notify the people they're used on. And deepfakes — plus AI-written text on matters of public interest — must be labeled as artificial.

If a customer can interact with your AI or read what it wrote, you now carry a disclosure duty — and the burden of proving you disclosed sits with you, not the regulator.

There is one partial reprieve, and it's easy to misread. The technical watermarking of synthetic content already on the market before Sunday gets a four-month grace period, to December 2.

The duty to tell someone they're talking to a bot does not get that grace — that one is live immediately.

What to do about it

You don't need a compliance department to clear this bar. Most of the work is plain disclosure you can add in an afternoon:

  • Find where AI meets an outsider. Walk your customer touchpoints: the website chatbot, the voice system, the email assistant that hits "send," any AI images or copy you publish.
  • Add the sentence. A plain line — "You're chatting with an AI assistant" — satisfies the direct-interaction rule. It costs nothing, and it's the thing a customer or a regulator can check in a single click.
  • Label what your AI makes. Mark AI-generated media and public-facing text as artificial, and keep it visible rather than buried in a policy page.
  • Push it onto your vendors. If a third party supplies your chatbot or content tool, get the disclosure and watermarking commitments in writing — you're the "deployer," so the duty lands on you regardless of who built it.

The parts of the AI Act that need lawyers were delayed. The part that needs a sentence was not — and Sunday is the day "tell people it's AI" stops being a courtesy and becomes the law.

Notes & sources European Commission, Shaping Europe's digital future: the official AI Act timeline. The regulation entered into force on August 1, 2024 and applies in phases — prohibited practices from February 2025, general-purpose AI and governance rules from August 2, 2025, and the next major tranche of obligations from August 2, 2026. EU Artificial Intelligence Act, Article 50: sets out the transparency obligations that apply from August 2, 2026 across four situations — direct interaction with an AI system, machine-readable marking of synthetic audio/image/video/text, notification for emotion-recognition and biometric-categorisation systems, and disclosure of deepfakes and AI-generated text on matters of public interest. Article 50 obligations are not limited to "high-risk" systems; under Article 99 the Act sets penalties for such breaches at up to €15 million or 3% of total worldwide annual turnover, whichever is higher. Sidley, Data Matters (Jun 24, 2026): a practical guide to preparing for the August 2, 2026 transparency deadline, confirming that providers must disclose AI interaction and mark synthetic outputs while deployers must label deepfakes and public-interest text. Notes a four-month grace period — to December 2, 2026 — that applies only to the content-marking (watermarking) obligation for systems placed on the market before August 2, not to the other transparency duties. Gibson Dunn: analysis of the EU AI Act "Omnibus" agreement, which postponed the high-risk obligations for stand-alone Annex III systems to December 2, 2027 and for AI embedded in regulated products to August 2, 2028, while confirming that the Article 50 transparency obligations largely remain on their original August 2, 2026 schedule.
Signal check

Also worth knowing today

AI News

California's disclosure rule lands the same day — on purpose.

The California AI Transparency Act becomes operative August 2, 2026, a date moved from January specifically to line up with Europe's. Providers of large public generative-AI systems must offer a free AI-detection tool and embed disclosures in what they generate, with penalties of $5,000 per violation per day. The operator read: "say when it's AI" is now arriving on both sides of the Atlantic in the same week.

Aug 2, 2026Source →
AI News

Most enterprises still can't make AI pay back what they spend on it.

A survey of 639 senior enterprise AI leaders found 57% say their AI returns still fail to outpace their AI spend — unchanged since 2025 — even as 93% report improved production capability. The signal beneath the compliance news: the cost of getting AI right, including the boring parts like disclosure, only pays off when the tool is aimed at a decision that actually moves money.

Jul 21, 2026Source →