BRUSSELS — New European Union transparency requirements for artificial intelligence took effect Sunday, requiring companies to tell people when they are interacting with certain AI systems and to mark some computer-generated or manipulated content.
The provisions are part of the EU’s AI Act, a broad regulatory framework adopted in 2024 and scheduled to take effect in stages. The latest rules apply to providers and deployers of AI systems operating in the 27-nation bloc, including companies based outside Europe that offer services to European users.
Under guidance published by the European Commission, providers must design certain systems to inform people when they are interacting directly with AI. Providers of systems that generate or alter text, images, audio or video must also make those outputs detectable through machine-readable marking.
Additional disclosure requirements apply to deepfakes, AI-generated material about matters of public interest that lacks human review or editorial control, and systems used for emotion recognition or biometric categorization.
New rules for products and content
The requirements are intended to help users distinguish between human-created and synthetic material. They affect product interfaces, content-generation systems and the processes companies use to publish or distribute AI-assisted media.
The rules do not require every use of AI to carry the same label. Obligations vary according to the type of system, the role of the company using it and the way the output reaches the public. The Commission’s guidelines describe the scope of the rules and the responsibilities assigned to providers, which build or supply AI systems, and deployers, which use them.
The Commission has also developed a voluntary code of practice intended to help companies implement consistent methods for marking and labeling generated content. The legal requirements, however, are established by the AI Act rather than the voluntary code.
For publishers, advertisers and brands, the provisions place particular attention on public-facing synthetic media. Human review and editorial control are relevant to whether certain public-interest text requires a disclosure, while deepfakes are subject to separate labeling rules.
Enforcement begins as the rulebook changes
The EU is expanding its AI Office, the body responsible for overseeing parts of the law. The Associated Press reported that 38 additional staff members will monitor companies and that the Commission has opened confidential whistleblower and compliance channels.
The Commission can investigate potential violations, request documentation and impose penalties or restrict access to the European market when companies fail to comply with applicable provisions.
The transparency rules arrived days after a separate measure adjusted other parts of the regulatory schedule. The AI Omnibus, which took effect July 27, extended deadlines for some high-risk systems, expanded access to regulatory testing programs and simplified obligations for smaller companies. It also expanded the AI Office’s oversight of certain systems built on general-purpose models.
Under the revised schedule, rules for some high-risk systems begin in December 2027, while requirements for high-risk AI embedded in regulated physical products begin in August 2028. The transparency duties that began this week were not postponed.
A test of implementation
The new phase will test whether disclosures can remain clear across products that generate or alter content in different formats. It will also test whether machine-readable marks remain detectable when material moves among platforms or is edited after creation.
Companies serving the European market are expected to review user notices, content-labeling practices, technical metadata and internal records to determine which systems fall within the rules.
The Commission has said the goal is to reduce deception and manipulation while allowing continued development of AI products. The practical effect will depend on how companies implement the requirements and how regulators interpret them during investigations.