Introduction
Many thanks for intern Bo-Rui Hao for her assistance in preparing this blog.
As artificial intelligence becomes increasingly embedded in products, services and business operations, regulators are focusing not only on safety and accountability, but also on AI transparency. One of the core objectives of the EU AI Act (which we have covered in a previous blog) is to ensure that individuals understand when they are interacting with AI systems and when content has been artificially generated or manipulated.
Article 50 of the EU AI Act introduces a series of transparency obligations that apply to both providers and deployers of AI systems. These requirements are designed to enhance trust, reduce deception and support informed decision-making by users and consumers. Importantly, many of these obligations will apply across a wide range of AI use cases within the EU, including customer service chatbots, generative AI tools, deepfakes, emotion recognition systems and AI-generated public communications.
Who Must Adhere to the AI Transparency Obligations?
As, above different transparency obligations apply to providers and deployers.
Under Article 3 of the EU AI Act, a ‘provider’ is a person or entity that develops any AI system and places it on the EU market, and a ‘deployer’ is anyone person or entity that uses an AI system commercially and / or professionally.
Therefore, even if you are developing or using your AI model through your UK-registered company, if you are distributing it in the EU or implementing it as part of your operations in the EU, then these obligations may apply to you.
That said, these specific transparency obligations are relevant only for specific types of AI models:
- AI systems that interact with humans;
- AI systems generating synthetic content such as synthetic audio, video, or images;
- AI systems for emotion recognition or biometric categorisation systems;
- AI systems generating deepfake image, audio or video content;
- AI systems generating or manipulating text that is of public interest.
AI Systems That Interact with Humans
One of the most widely applicable obligations concerns AI systems intended to interact directly with natural persons.
Under Article 50(1), providers (not deployers) must ensure that AI systems are designed and developed so that individuals are informed that they are interacting with an AI system. This applies to systems such as:
- Customer service chatbots;
- Virtual assistants;
- AI-powered support agents;
- Automated recruitment assistants;
- AI-generated customer communications; and
- Conversational AI embedded in websites or applications.
The obligation applies except in the following circumstances:
- It is obvious to a reasonably well-informed, observant and circumspect person that they are interacting with AI, taking into account the context and circumstances of use.
- The AI model is authorised by law to detect, prevent, investigate, or prosecute criminal offences (provided the system is not publicly).
AI Systems Generating Synthetic Content
Under Article 50 (2), providers (not deployers) of AI systems that generate synthetic audio, image, video or text content must ensure that outputs are marked in a machine-readable format and are detectable as artificially generated or manipulated.
The obligation applies broadly to:
- Text generation systems;
- Image generation tools;
- Video generation platforms;
- Voice cloning applications; and
- Other systems capable of producing synthetic content.
The legislation does not prescribe a specific technical solution. Instead, providers must implement measures that are:
- Effective;
- Interoperable;
- Robust; and
- Reliable,
to the extent that doing so is technically feasible.
In reality, this article will impose obligations on all generative AI systems (e.g. ChatGPT) to implement measures ensuring that all AI outputs are suitably marked. There has been much discussion around how AI developers can best meet his obligation. One potential solution is to brand each generated output with a watermark indicating the specific AI system which was used to generate it. Another option is to include digitally signed and time-stamped metadata recording the presence of AI generated data. These options have their pros and cons, and their effectiveness will vary with each particular AI system and circumstance.
In assessing compliance, organisations are expected to consider factors such as:
- The type of content involved;
- Technical limitations;
- Implementation costs; and
- The current state of technological development and relevant standards.
Importantly, there are exceptions, namely where:
- The AI system performs merely an assistive function for standard editing conducted by a human; or
- The AI system in question is authorised by law to detect, prevent, investigate or prosecute criminal offences.
AI Systems for Emotion Recognition or Biometric Categorisation Systems
Under Article 50(3), individuals exposed to such systems must be informed of their operation by the deployer (not providers).
These technologies are particularly sensitive because they seek to infer emotions, characteristics or categories from individuals based on biometric data.
Examples may include systems used to:
- Assess emotional responses;
- Analyse customer engagement;
- Categorise individuals based on biometric characteristics; or
- Monitor behavioural indicators.
In addition to providing notice, deployers must ensure that any personal data processing complies with applicable data protection legislation, including the General Data Protection Regulation (GDPR).
For many organisations, compliance with Article 50(3) will therefore require a combined AI governance and privacy compliance approach. Transparency notices, privacy information, lawful basis assessments and data protection impact assessments may all become relevant considerations.
This obligation applies except in the instance where the AI system is permitted by law to detect, prevent, or investigate criminal offences, subject to appropriate safeguards being put in place.
AI Systems Generating Deepfakes
Under Article 50(4), deployers (not providers) of AI systems that generate or manipulate image, audio or video content constituting a deepfake must disclose that the content has been artificially generated or manipulated.
The provision is intended to address increasing concerns about misinformation, impersonation and reputational harm arising from synthetic media.
The obligation applies to organisations and individuals deploying AI-generated content as well, rather than only the providers of the underlying AI tools. This distinction is important because compliance responsibilities may arise downstream in the AI value chain.
For example, a marketing team using generative AI to create synthetic videos or voiceovers may itself have disclosure obligations even where the AI provider has complied with its own obligations under the Act.
The AI Act recognises, however, that disclosure requirements should not unnecessarily interfere with legitimate creative expression.
Where deepfake content forms part of an evidently:
- Artistic;
- Creative;
- Satirical;
- Fictional; or
- Similar work,
the transparency obligation is limited to disclosing the existence of AI-generated or manipulated content in an appropriate manner that does not undermine the display or enjoyment of the work.
This obligation applies except in the instance where the AI system is permitted by law to detect, prevent, or investigate criminal offences.
AI Systems Generating or Manipulating Text that is of Public Interest
Article 50(4) also requires deployers (NOT providers) of AI systems that generate or manipulate text intended to inform the public on matters of public interest to disclose that the text has been artificially generated or manipulated.
This provision is particularly relevant for:
- News publishers;
- Media organisations;
- Public bodies;
- Political communications; and
- Organisations publishing information intended to shape public understanding of important issues.
Search engines and tools which rely on up to date data to generate their output may also fall within scope of this article.
The objective is to ensure that readers are aware when AI has played a significant role in producing information that may influence public opinion or decision-making.
However, the obligation contains important exceptions, namely where:
- The AI-generated content has undergone a process of human review, where a natural or legal person holds editorial responsibility for the publication of the content; or
- The AI system in question is authorised by law to detect, prevent, investigate or prosecute criminal offences.
Accessibility and Timing Requirements for AI Transparency Obligations
The AI Act does not merely require transparency; it specifies how transparency information must be delivered.
Under Article 50(5), information provided under the transparency obligations must be:
- Clear;
- Distinguishable; and
- Provided no later than the first interaction or exposure.
In addition, the information must comply with applicable accessibility requirements.
This means organisations should consider not only the content of disclosures but also their presentation. Hidden notices, obscure wording or inaccessible formats may not satisfy the requirements.
Businesses should therefore review user interfaces, onboarding processes and content publication workflows to ensure disclosures are both visible and understandable.
Timescales
The AI transparency obligations for both providers and deployers will come into legal effect on 2 August 2026. Providers and deployers therefore need to understand what their obligations are, and ensure that they are compliant with this legislation before that date.
It is important to note that if your system is already on the market, you will not need to meet the obligation under Article 50 (2) (i.e. to ensure that outputs are marked in a machine-readable format and are detectable as artificially generated or manipulated) until 2 December 2026, as per the AI Omnibus provisional agreement that took place in May this year.
While these obligations apply from 2 August 2026, the AI Omnibus provisional agreement of May 2026 grants generative AI systems already on the market before that date until 2 December 2026 to meet the machine-readable marking requirement under Article 50(2).
Looking Ahead: Codes of Practice and Future Standards for AI Transparency Obligations
Recognising the technical challenges associated with content labelling and detection, Article 50(7) empowers the AI Office to encourage the development of voluntary codes of practice at EU level.
These codes are intended to facilitate consistent implementation of obligations relating to the detection and labelling of AI-generated content.
The European Commission may formally approve such codes through implementing acts. If existing industry-led initiatives are deemed inadequate, the Commission may adopt binding common rules governing implementation.
As a result, organisations should expect the transparency framework to continue evolving over the coming years. Technical standards, watermarking solutions and content authenticity mechanisms are likely to become increasingly important elements of AI compliance programmes.
How Businesses Should React to the AI Transparency Obligations
The transparency obligations under the EU AI Act represent a comprehensive attempt to regulate how AI systems and AI-generated content are disclosed to individuals.
While the requirements may appear straightforward, their practical implementation raises significant operational, legal and technical questions. Businesses will need to determine when AI interactions require disclosure, how synthetic content should be labelled, what constitutes adequate editorial control, and which technical solutions can satisfy machine-readable marking requirements.
Organisations developing or using AI systems should begin assessing their existing AI use cases now. Mapping AI systems, reviewing customer-facing interactions, evaluating content generation processes and strengthening governance frameworks will all be important steps towards compliance.
Ultimately, the transparency provisions are designed to ensure that individuals can recognise when AI is involved and make informed decisions accordingly. For businesses, achieving compliance will not simply be a regulatory exercise – it will also be an opportunity to build trust and demonstrate responsible AI governance in an increasingly AI-driven economy.
At EM Law, we are at the cutting edge of legal developments in this area and have assisted many clients in the UK and internationally with developing and deploying AI systems. If you have any questions about AI or AI compliance, please don’t hesitate to contact us here, or visit our AI Lawyers and Software & Tech Lawyers pages for more information.
