The EU’s new AI transparency rules: What businesses need to know

AI is becoming part of everyday work. It writes, edits, responds, generates images and helps people make sense of information. Often, that’s useful. Sometimes, though, it’s difficult to know whether we are dealing with a person, a machine or a bit of both.

From 2 August 2026, new transparency obligations under Article 50 of the EU AI Act will make some of those distinctions clearer. The aim is simple: people should know when they are interacting with AI or viewing content that has been generated or manipulated by it.

The European Commission has now published guidelines explaining how the rules apply. Here are the main points for organisations using or providing AI systems.

Providers and deployers: What’s the difference?

The AI Act separates organisations into two groups.

A provider develops an AI system, or has one developed, and places it on the EU market or puts it into service under its own name or trademark. Providers based outside the EU may also fall within the rules where their system’s output is used in the EU.

A deployer uses an AI system under its authority in a professional context. This might be a company using generative AI to create campaign materials.

That difference matters because providers and deployers have different responsibilities.

When people are talking to AI

Providers of directly interactive AI systems, including chatbots, AI agents and avatars, must design them so people are informed that they are interacting with AI, unless it is already obvious.

The information should be clear and distinguishable, meet accessibility requirements and appear from the beginning of the first interaction. AI working entirely in the background, communicating only with another machine or operating through a human intermediary is not covered by this particular obligation.

In practical terms, businesses using a customer-facing chatbot should check how well it works, and how clearly it introduces itself.

Marking AI-generated content

Providers of generative AI systems must make AI-generated or manipulated audio, images, video and text detectable using machine-readable marking.

Not every use of AI triggers this requirement. The Commission’s guidelines include exceptions for standard editing, such as spelling and grammar corrections. Source code, short sequences of symbols or numbers, some machine-to-machine outputs and certain closed-loop industrial uses may also fall outside its scope. The important distinction is between AI assisting with ordinary editing and AI generating or manipulating the content itself.

What do the EU’s new AI transparency rules mean for my business?

The same organisation can be a provider, a deployer or both, depending on how it uses AI. Here are three practical examples.

You are the chief communications officer at an Irish university or third-level institution

If your team uses AI to draft a LinkedIn post, prospectus copy or student information, the institution is acting as a deployer. You do not need to label every AI-assisted draft. However, AI-generated text about matters of public interest may require disclosure unless it receives meaningful human review, fact-checking and editorial sign-off. Realistic synthetic images or videos that could appear authentic should carry a clear label in the post, caption or printed material.

You are the founder of a SaaS or technology company

If you use AI to create website copy, LinkedIn posts or product videos, you are acting as a deployer for those marketing activities. Ordinary AI-assisted marketing does not automatically require a public label, but realistic synthetic footage or deepfake-style content should be disclosed clearly. For example, in a TikTok caption or beside a video on your website. If your SaaS product itself uses AI, you may also be a provider and have additional responsibilities.

You are a food producer working with B2B vendors

If you use AI to create social images, short videos, LinkedIn posts or printed marketing materials for B2B clients, you are likely acting as a deployer. You do not need to label every AI-assisted draft, but if the final asset is realistic enough to look like genuine photography or footage, a visible disclosure may be needed when it is published. Make that decision part of your approval process, and agree in advance who is responsible for adding the disclosure, whether that is in the caption, on the video itself or alongside the printed piece.

Labelling deepfakes and public-interest content

Deployers have a separate responsibility to disclose certain AI-generated material in a way people can understand without special tools.

This includes deepfakes: AI-generated or manipulated images, audio or video that resemble real or plausibly real people, objects, places, entities or events and could falsely appear authentic. A machine-readable mark alone is not enough. The disclosure must be clear and perceivable by the audience no later than their first exposure to the content.

The rules also cover AI-generated or manipulated text published to inform the public about matters of public interest. This can include politics, public services, health, safety, environmental protection and economic, scientific or cultural developments relevant to public debate.

There is an important exemption. Text does not need this label where it has undergone meaningful human review or editorial control and a person or organisation holds editorial responsibility for its publication. A quick spell-check or purely procedural approval does not count. The review must consider the substance of the work, including its accuracy and the trustworthiness of its sources.

Deployers must also inform people exposed to emotion-recognition or biometric-categorisation systems.

Who decides on AI disclosure?

There is no single GDPR-style “AI disclosure officer.” In most organisations, the decision on whether a disclosure is needed should sit with the person responsible for final content sign-off, usually in communications, marketing or editorial, with legal or compliance input where necessary. That person should decide whether the content is text, image, video or another format, whether AI was used only to assist or materially generate it, whether it is public-facing, and whether a visible disclosure should be added before publication. For higher-risk content, it helps to name one internal owner for AI transparency so there is a clear point of accountability.

A simple internal record is usually enough, noting what content was created, what AI tools were used, who reviewed it, and whether a disclosure was added. That record can be a spreadsheet, a checklist or a short approval log. The aim is to make the decision traceable, not bureaucratic, in case of an audit.

Key takeaways

Article 50 applies from 2 August 2026. A limited grace period applies to AI systems placed on the market before that date, their providers have until 2 December 2026 to meet the machine-readable marking and detection requirements. Content generated before 2 August does not have to be labelled retrospectively.

Enforcement will sit mainly with national market-surveillance authorities, with specific roles for the EU AI Office and the European Data Protection Supervisor. Fines can reach €15 million or 3% of worldwide annual turnover, with proportionality taken into account for smaller businesses.

For most organisations, the useful place to begin is not with a new policy document, but with a simple audit. Where are we using AI? Who sees its outputs? Who checks the substance of what we publish? And would an ordinary person understand when AI is involved?

Transparency does not make creative or useful AI less valuable. Done properly, it gives people the context they need to trust it. Piquant will be monitoring this space closely.

Written by Pawel Bogucki, Head of Development at Piquant.

This article is intended as an informative overview of the EU AI Act transparency rules. It is not legal advice.

If you are interested in optimising your business for digital transformation, reach out for a chat.