The EU’s New AI Labels Begin Today: Deepfakes and Chatbots Must Reveal Themselves

From today, artificial intelligence operating in the European Union becomes harder to hide behind the appearance of human communication.
On 2 August 2026, major transparency provisions under the EU AI Act became applicable. The rules require people to be informed when they are directly interacting with certain AI systems, while providers of generative systems must make AI-generated or manipulated content detectable through machine-readable marking. Deepfakes and some AI-generated text concerning matters of public interest must also carry visible disclosures. �
Digital Strategy +2
The rules arrive at a time when synthetic voices, photographs and videos can appear convincing enough to influence consumers, voters and news audiences before verification catches up.
What Must Be Labelled?
The law does not simply require a large “Made by AI” banner on everything touched by software.
Different obligations apply to different people and systems.
Providers of generative AI must design their systems so that synthetic or manipulated outputs can be identified in a machine-readable form. People must also be informed when they are communicating directly with an AI system unless that fact is already obvious from the context. �
Digital Strategy +1
Visible labels are particularly relevant for:
Deepfake images, video and audio
AI-generated or manipulated public-interest material
Chatbots presented in ways that could be mistaken for humans
Synthetic content designed to resemble authentic events or people
There are exceptions, including some artistic, satirical and human-edited material. Content created before 2 August 2026 does not have to be labelled retrospectively, although the European Commission encourages voluntary disclosure where practical. �
Digital Strategy +1
Why the Rules Matter
The internet’s old question was: “Is this claim true?”
The AI era adds another question: “Did the person, voice or event exist at all?”
A realistic video can be shared millions of times before a correction reaches the same audience. A synthetic customer-service representative may appear human enough that users disclose sensitive information without understanding who—or what—is receiving it.
Transparency does not make every AI output trustworthy. It gives people essential context for deciding how much trust to place in what they see.
Will Labels Solve the Deepfake Problem?
Not entirely.
A dishonest creator may remove a visible label. Content can be cropped, recorded from another screen or uploaded to platforms that do not preserve provenance information.
That is why the EU approach includes both human-visible disclosure and machine-readable marking. Platforms and verification systems need technical signals that can travel with the file even when ordinary viewers cannot see them. �
Digital Strategy +1
However, no watermark is indestructible. The system will depend on cooperation among AI developers, social platforms, publishers, regulators and users.
The New Enforcement Team
The European Commission’s AI Office and national authorities are now responsible for enforcing the applicable rules. The EU has expanded its Brussels enforcement operation and introduced channels for reporting suspected violations. �
AP News +1
The timing is significant. European officials are also speaking with OpenAI and Anthropic following recent cases in which advanced AI agents accessed real external systems during cybersecurity testing. �
Reuters
AI transparency and AI safety are therefore becoming part of the same policy conversation.
Mary Chuks’ Perspective
Labels will not remove the need for critical thinking.
A clearly labelled AI video may still contain useful education, brilliant art or accurate information. An unlabelled human video may still be deceptive.
The label answers only one question: how the content was produced or altered.
Users must still ask:
Who published it?
What evidence supports it?
Was it edited?
Does the context match the original event?
Is the emotional reaction being deliberately manipulated?
Responsible AI is not achieved by hiding artificial intelligence. It is achieved by allowing people to understand when and how it is participating.
Practical Takeaways for Creators
Creators publishing for European audiences should:
Clearly identify realistic AI-generated or manipulated media.
Preserve platform provenance and watermarking features.
Avoid presenting conceptual AI artwork as documentary evidence.
Disclose AI-generated public-interest text where legally required.
Review whether their chatbot adequately informs users that it is automated.
Keep records showing how content was produced and edited.
Conclusion
The EU has made a fundamental choice: when AI imitates reality, people deserve to know.
The success of the law will not be measured merely by the number of labels appearing online. It will be measured by whether those labels help people pause long enough to distinguish artificial appearance from reliable evidence.
Original Sources and Further Reading
European Commission: “Commission Starts Enforcing AI Act Rules and New Transparency Requirements on 2 August,” published 31 July 2026. �
Digital Strategy
European Commission guidance: Transparency obligations for generative and interactive AI systems under Article 50. �
Digital Strategy +1
Independent reporting: The Guardian, “AI Labels to Be Compulsory on Authentic-Looking Content Under EU Rules.” �
The Guardian