The EU AI Act: What Falls Outside the New Transparency Rules

New European Union transparency rules for artificial intelligence took effect on 2 August 2026. They aim to prevent people from being deceived by chatbots, deepfakes and certain forms of AI-generated public information. They do not prevent ordinary businesses from using AI to save time, improve communication or develop ideas.

Tools such as OpenAI’s ChatGPT, Google’s Gemini, Meta AI on Facebook and xAI’s Grok can still help businesses with writing, research, summaries and everyday administration. The important question concerns how the business uses the finished work.

Private assistance remains private

An architect may use ChatGPT to turn rough notes into a clear letter to a client.

A coffee-shop owner may ask Google Gemini to summarise staff suggestions, prepare a draft training guide or suggest names for a new cake.

A village postmistress may use Grok to simplify a supplier’s email or use Meta AI on Facebook to help draft a friendly post about a shop promotion.

These people do not generally need to announce that AI helped them. The customer still deals with a human being, and that human remains responsible for checking and approving the result.

The chatbot disclosure rule applies when the AI itself communicates directly with a person. AI working privately in the background falls outside that particular requirement.

Human editing makes an important difference

The rules pay particular attention to AI-generated text published to inform the public about matters such as politics, health, law, public services, safety, finance and scientific developments.

However, such text does not require an AI label when a knowledgeable person has properly reviewed it, checked its claims and accepted editorial responsibility for it.

A quick spell-check does not count as meaningful review. A human editor needs to examine the substance, sources and accuracy of the material.

This means a business can use AI to prepare a first draft of an article, report or public notice. A suitable person should then:

• Check the facts.

• Correct misleading statements.

• Remove invented information.

• Approve the final version.
  
  

Where that level of review has taken place, the law does not require a label merely because AI helped to produce the first draft.

Ordinary editing remains acceptable

The regulations also distinguish between creating deceptive synthetic content and using AI for standard editing.

Routine assistance may include correcting spelling, improving grammar, adjusting formatting, cleaning up sound or making modest technical changes that do not substantially alter the meaning of the original material.

The European Commission also identifies source code, short strings of numbers or symbols, machine-to-machine messages and some closed production work as falling outside the machine-readable marking requirement.

Background business systems usually need no announcement

A customer does not need to receive an AI warning merely because a business uses software to:

• Filter spam.

• Forecast stock requirements.

• Examine sales patterns.

• Suggest staff rotas.

• Organise files.

• Detect duplicate records.

• Plan deliveries.

• Summarise internal meetings.
  
  

These systems work behind the scenes rather than holding a direct conversation with the public. The European Commission specifically states that background systems and machine-to-machine communication fall outside the chatbot disclosure duty.

Other laws, including data-protection, employment and consumer-protection rules, may still apply. “Outside the transparency rule” does not mean “outside every law”.

An imaginary image is not automatically a deepfake

Suppose a coffee company uses AI to create an illustration of an imaginary café beside a lake. That image does not necessarily amount to a deepfake.

A deepfake resembles a real or plausibly real person, place, object or event and could falsely appear authentic. A fabricated photograph showing a real architect approving a building that they have never seen would present a different problem and should carry a clear disclosure.

Providers of generative AI systems carry responsibilities for placing machine-readable information in synthetic content. Businesses should preserve such information rather than stripping it from files.

A visible label becomes especially important when an image, recording or video could mislead people into believing that a real event took place.

Personal use also sits outside the Act

The AI Act excludes purely personal, non-professional activity.

A person creating an amusing picture for a family chat does not become a business “deployer” merely by using AI. The position changes when someone uses the technology regularly for commercial, freelance or professional gain.

Where businesses should give a clear warning

A business should provide a clear notice when a customer speaks directly to an AI assistant.

A simple introduction will usually serve the purpose:

“Hello. I am an AI assistant. I can answer common questions and help you contact a member of our team.”

Businesses should also disclose deepfakes and take particular care with emotion-recognition or biometric-categorisation systems.

The safest principle remains simple:

Use AI freely as an assistant. Keep a responsible human in command. Tell people when they are dealing directly with a machine or could mistake synthetic material for reality.

Download the comprehensive guide

This article offers a brief introduction. Our downloadable guide, A Plain-English Guide to the EU AI Act for Small Businesses, explains the rules in greater detail.

It includes practical examples for an architect’s practice, a chain of coffee shops and a village post office, together with suggested notices and a simple business checklist.

This article provides general information rather than legal advice. Businesses using AI in sensitive, regulated or high-risk areas should seek appropriate professional guidance.