Does every AI chatbot need an "I am an AI" notice?

Yes, unless it is clear from the context that the user is interacting with an AI - for example a product explicitly branded as "AI Assistant." Where there is any ambiguity about whether the user understands they are talking to an AI, a disclosure is required. The obligation falls on the provider of the chatbot system.

What does "machine-readable marking" for AI-generated content mean?

It means embedding metadata or a signal in the output file that allows automated systems (not just humans) to detect that the content is AI-generated or manipulated. Standards are still being developed by the Commission and standardisation bodies. Existing technologies like C2PA watermarking are candidate approaches.

Are Article 50 violations subject to the same fines as high-risk non-compliance?

Yes - Article 50 violations fall under the Tier 2 penalty: up to €15 million or 3% of worldwide annual turnover. This is the same tier as non-compliance with other obligations, including high-risk requirements.

Do these transparency obligations apply to internal AI tools?

The chatbot disclosure obligation applies where a natural person is interacting with the AI system - so an internal employee-facing AI assistant is covered. The emotion-recognition and biometric-categorisation notification applies whenever persons are subject to the system, including employees. AI-generated content marking mainly concerns outputs disseminated externally, but the provider obligation attaches to the system regardless.

Can you use satirical deepfakes without restriction?

The artistic/satirical carve-out does not remove the disclosure requirement - it only permits a different form: the content must be "clearly labelled" as artificially generated or manipulated. A deepfake political satire video is allowed as long as it carries visible labelling; it cannot simply be published without any disclosure of its AI-generated nature.

EU AI Act topic guide

Transparency obligations under the EU AI Act (Article 50)

Not every AI use is banned or strictly regulated - many AI systems sit in a "limited/transparency risk" tier where the main obligation is to be honest about the AI's role. Article 50 sets out four distinct transparency duties covering chatbots, AI-generated content, deepfakes and emotion/biometric categorisation systems. These obligations apply from 2 August 2026 for most providers.

Reviewed by the AI Act Navigator team · Last updated 9 June 2026

TL;DR

Chatbots and conversational AI: users must be told they are interacting with an AI system - unless it is obvious.
AI-generated/synthetic content: providers must mark outputs (audio, image, video, text) in a machine-readable format as artificially generated or manipulated.
Deepfakes: deployers must disclose when content has been artificially generated or manipulated - with limited carve-outs for artistic/satirical works with appropriate labelling.
Emotion recognition and biometric categorisation: deployers must inform the people being processed.
[OMNIBUS - PROPOSED] The Digital Omnibus provisional agreement would reduce the grace period for AI-content marking to 3 months (effective 2 December 2026 instead of 2 August 2026).

Classify your AI system Obligations guide

Scope

What this covers

Chatbots and conversational AI systems (Article 50(1)): providers must ensure that natural persons interacting with AI are informed they are interacting with an AI system - unless it is evident from context or the system is used for criminal-investigations or similar lawful purposes.
AI-generated content - machine-readable marking (Article 50(2)): providers of AI systems generating synthetic audio, image, video or text content must ensure outputs are machine-readably marked as artificially generated or manipulated. Technical standards to be developed.
Deepfakes - deployer disclosure (Article 50(4)): deployers who use AI to generate or manipulate content showing real persons, places or events that falsely appears authentic must disclose the artificial nature. Carve-out for legitimate artistic, creative or satirical works, provided the content is clearly labelled.
AI-generated text for public information (Article 50(3)): deployers publishing AI-generated text to inform the public on matters of public interest must disclose the AI-generated nature - unless this text has undergone substantial human editing.
Emotion recognition and biometric categorisation (Article 50(5)): deployers of AI systems that recognise or infer emotions, or categorise natural persons based on biometric data, must inform the persons concerned in advance.

Article 50 transparency obligations apply from 2 August 2026 for providers (subject to the proposed Digital Omnibus adjustment for AI-content marking). They are separate from and lighter than the high-risk obligations. Many AI systems will only be subject to Article 50, not to Chapter III high-risk requirements.

Source: Regulation (EU) 2024/1689 (EUR-Lex)

Compliance challenges

Key compliance challenges

Machine-readable marking standards for AI-generated content are still being developed - providers cannot yet buy a simple technical solution off the shelf.
The "obvious AI" exception for chatbots is narrow: where a system is branded as an AI assistant, users are likely informed; but where AI is embedded in a human-seeming interface, disclosure is required.
Deepfake disclosure in creative/satirical contexts requires clear labelling - the artistic carve-out does not mean no disclosure, only that a different labelling form is acceptable.
Emotion recognition notification must be given "in advance" to the persons concerned - this can create operational challenges in contexts like retail or transport.

The EU AI Act applies a risk-based approach: obligations scale with the level of risk posed. AI Act high-level summary

What to do

Identify every AI system that has a conversational or interactive interface: implement a visible AI-disclosure notice for users.
Audit AI systems that generate synthetic audio, image, video or text: prepare for machine-readable marking when the technical standards land.
Review any use of AI to generate content depicting real people, places or events: implement deepfake disclosure where the content could mislead.
Identify emotion-recognition or biometric-categorisation deployments: design advance notification to affected persons.
Monitor the Digital Omnibus adoption for possible changes to the AI-content marking timeline.

For the full obligations breakdown, see the AI Act obligations guide, and for role-specific duties see the provider vs deployer guide.

[PROPOSED] Digital Omnibus changes to Article 50 - not yet law

The provisional Digital Omnibus political agreement of 7 May 2026 would reduce the grace period for AI-generated content marking (Art. 50(2)) from 6 months to 3 months, making the effective date 2 December 2026 rather than 2 August 2026. This is not yet formally adopted. Until adoption, the 2 August 2026 deadline in current law remains binding. Council press release, 7 May 2026

FAQ

Transparency obligations: common questions

Does every AI chatbot need an "I am an AI" notice?: Yes, unless it is clear from the context that the user is interacting with an AI - for example a product explicitly branded as "AI Assistant." Where there is any ambiguity about whether the user understands they are talking to an AI, a disclosure is required. The obligation falls on the provider of the chatbot system.
What does "machine-readable marking" for AI-generated content mean?: It means embedding metadata or a signal in the output file that allows automated systems (not just humans) to detect that the content is AI-generated or manipulated. Standards are still being developed by the Commission and standardisation bodies. Existing technologies like C2PA watermarking are candidate approaches.
Are Article 50 violations subject to the same fines as high-risk non-compliance?: Yes - Article 50 violations fall under the Tier 2 penalty: up to €15 million or 3% of worldwide annual turnover. This is the same tier as non-compliance with other obligations, including high-risk requirements.
Do these transparency obligations apply to internal AI tools?: The chatbot disclosure obligation applies where a natural person is interacting with the AI system - so an internal employee-facing AI assistant is covered. The emotion-recognition and biometric-categorisation notification applies whenever persons are subject to the system, including employees. AI-generated content marking mainly concerns outputs disseminated externally, but the provider obligation attaches to the system regardless.
Can you use satirical deepfakes without restriction?: The artistic/satirical carve-out does not remove the disclosure requirement - it only permits a different form: the content must be "clearly labelled" as artificially generated or manipulated. A deepfake political satire video is allowed as long as it carries visible labelling; it cannot simply be published without any disclosure of its AI-generated nature.

Get AI Act-ready

Use the risk classifier to find your system's tier, then explore the obligations and checklist for your role.

Classify your AI system

This is guidance, not legal advice

This is guidance to help you understand how the EU AI Act applies to transparency obligations, not legal advice. For decisions specific to your organisation, confirm with the official sources we link or a qualified legal adviser.

Sources

[1]Regulation (EU) 2024/1689 (EU AI Act) - EUR-Lexretrieved 9 Jun 2026
[2]European Commission: AI regulatory frameworkretrieved 9 Jun 2026
[3]AI Act Explorer: high-level summaryretrieved 9 Jun 2026
[4]AI Act implementation timelineretrieved 9 Jun 2026
[5]Council of the EU: Digital Omnibus provisional agreement, 7 May 2026retrieved 9 Jun 2026

The AI Act Brief

Subscribe to The AI Act Brief

We watch Brussels so you don't. Plain-English EU AI Act updates, free.

No spam. Unsubscribe anytime.

categoryWhat this covers

priority_highKey compliance challenges

checklistWhat to do

helpTransparency obligations: common questions