Artificial intelligence is becoming part of everyday digital services, from chatbots and image tools to systems that recommend or generate content. As those tools become more common, users need to know when they are interacting with AI and when media has been generated or altered. The European Union’s AI Act has begun applying its transparency framework, and one important transition date is approaching: 2 December 2026 for certain marking and detection obligations affecting AI systems placed on the market before 2 August 2026.
What changed on 2 August 2026
The European Commission’s AI Act Service Desk says that from 2 August 2026, provisions on prohibited AI practices, transparency requirements for certain AI systems and rules for general-purpose AI models became enforceable. The transparency rules are intended to help people recognise when they are interacting with an AI system and when content has been generated or manipulated by AI. The exact obligation depends on the type of system and the role of the provider or deployer.
This is not a blanket rule that every digital feature must display the same label in every situation. The Act contains different requirements for different uses. For example, interactive systems designed to communicate with people have disclosure duties, while certain AI-generated or manipulated content, including deepfakes, is subject to transparency obligations. Organisations need to understand which provisions apply to the systems they provide or use.
Why 2 December matters
The Commission’s FAQ identifies a transition period for certain AI systems placed on the market before 2 August 2026. Their providers have until 2 December 2026 to comply with the marking and detection obligations under Article 50(2). This deadline is specific; it should not be confused with the start date for every AI Act obligation or with the later rules for high-risk systems.
For businesses, the immediate task is to review whether a system falls within the relevant category and whether the required technical measures are in place. That may involve identifying generated or manipulated outputs, ensuring that markings are robust enough to be detected and documenting how the system communicates its use of AI. Legal and technical teams should consult the current official guidance rather than relying on a simplified social-media summary.
Transparency is about user understanding
Labels are useful only when people can understand them. A chatbot disclosure should make it clear that the user is communicating with an AI system rather than a human representative. Content markings should help audiences recognise when an image, video or audio recording has been generated or manipulated. If a notice is hidden in a long policy document or written in language that users cannot interpret, it may fail to serve its practical purpose.
Transparency also supports trust in legitimate uses of AI. People may be comfortable using an automated assistant for a routine task if they know what it is and what it can do. Problems arise when an AI system is presented as a person, or when synthetic media is shared in a way that invites viewers to treat it as an unaltered record of events. Clear disclosure gives people more information with which to make decisions.
What organisations should review
Companies should begin with an inventory of AI systems used in products, customer service, marketing and internal workflows. For each system, identify the provider, the date it was placed on the market or put into service, the intended use and the type of output it creates. This helps determine which obligations may apply and where additional technical or contractual information is needed.
Teams should then review user interfaces, output labelling, documentation and supplier agreements. If a third-party tool generates images or text, the organisation should understand what information the provider supplies and whether the outputs can be identified as AI-generated. A policy that says “we use AI responsibly” is not a substitute for a working process that explains how the system is used and how users are informed.
What users should look for
For individuals, the new framework is a reason to pay attention to context. A label does not automatically mean that a piece of content is false, just as the absence of a visible label does not prove that content is authentic. Users should still check the source, compare important claims with reliable reporting and be cautious when an image or audio clip is used as evidence for a serious allegation.
It is also sensible to ask a service whether a response comes from an automated system when that distinction matters. In customer support, finance, health or public services, knowing whether a person or a machine is answering can affect how much detail a user shares and how they seek help when a problem needs human judgement.
Keep the timeline precise
The AI Act’s implementation includes several dates. Some prohibitions and AI-literacy obligations applied earlier; transparency and other provisions became enforceable from 2 August 2026; the 2 December 2026 date concerns specific transition arrangements for certain pre-existing systems; and high-risk rules have separate later timelines. Treating all of these as one deadline creates confusion for both businesses and the public.
The practical message is straightforward: identify the systems in use, check the applicable rule and make transparency visible where required. Further reading: European Commission AI Act Service Desk — enforcement dates and Navigating the AI Act.
