EU AI Transparency Rules Are Live: What Small Businesses Should Check

The European Union's new AI transparency obligations began applying on August 2, 2026. If your business serves people in the EU, uses a customer-facing chatbot, or publishes certain AI-generated media, this is a good week to review how that use is disclosed.

This is not a blanket rule that every AI-assisted sentence needs a warning label. The obligations depend on what the system does and whether a business is acting as a provider or a deployer of that system.

What changed

Article 50 of the EU AI Act covers four broad transparency situations. Providers of certain AI systems must design them so people are informed when they are interacting directly with AI. Providers of generative systems also face machine-readable marking requirements for AI-generated or manipulated content.

Businesses that deploy AI may need to inform people when they are exposed to emotion recognition or biometric categorisation, deepfakes, or AI-generated text about matters of public interest that was published without human review or editorial control.

The European Commission published final guidance in July to clarify which providers, deployers, and systems are in scope. Its plain-language quick facts page is a useful starting point.

There is one important distinction: the transparency rules are now applicable, but the EU's separate AI Omnibus extended several deadlines for high-risk AI systems. According to the Commission, Annex III high-risk rules now start December 2, 2027, while rules for high-risk AI embedded in regulated products start August 2, 2028.

What this means for your business

The practical question is not simply, "Do we use AI?" It is, "Where does AI touch a customer or public-facing piece of content, and what does that person reasonably understand?"

Start with these five checks:

1. List customer-facing AI. Include website chat, automated phone or voice agents, support tools, recommendation experiences, and any custom AI feature.
2. Record who supplied it. Note whether you built the system, bought it, or configured a third-party service. Your role affects which obligations may apply.
3. Review the customer experience. A person should not have to guess whether they are interacting directly with an AI system. Check the opening message, interface, and handoff to a human.
4. Audit synthetic media. Find AI-generated or materially manipulated audio, images, video, and public-interest text. Record how each item was reviewed and whether it is labelled or machine-readable.
5. Ask vendors for evidence. Request their EU AI Act documentation, supported disclosure settings, content-marking method, and the date those features became available. Save the answers with your vendor records.

A simple internal record to make today

Create a one-page inventory with these columns:

• Tool or system
• Business purpose
• Where a customer encounters it
• Provider or deployer role
• Current disclosure
• Human review step
• Vendor documentation link
• Owner and next review date

That inventory will not decide your legal obligations, but it gives your team and legal adviser something concrete to assess. It can also reveal customer-experience problems before they become compliance problems.

What remains uncertain

Application depends on the facts, including the system, audience, business role, and any relevant exception. The Commission's voluntary Code of Practice on Transparency of AI-Generated Content can help organisations demonstrate certain marking and labelling practices, but it does not replace the law or case-specific advice.

This article is a practical business briefing, not legal advice. Businesses serving EU users should review the Commission's final Article 50 guidelines and consult qualified counsel where the scope is unclear.

Sources

• European Commission: Article 50 transparency guidelines (published July 20, 2026)
• European Commission: Quick facts on transparency rules (updated July 29, 2026)
• European Commission: AI Omnibus enters into force (published July 27, 2026)
• EU AI Act Service Desk: enforcement powers for advanced models (applicable from August 2, 2026)