2026-08-02 · EU · in force now
From August 2, 2026, transparency duties (labeling AI interaction, marking synthetic content, disclosing deepfakes) and the penalty regime for GPAI providers are enforceable in the EU. The AI Office gains power to demand information, evaluate models, order mitigations, and issue fines.
What it requires
Organizations must make clear to users when they are interacting with AI, mark synthetic content as such, and disclose deepfakes. GPAI model providers face active enforcement: the AI Office can request information about models, conduct evaluations, require specific mitigations to address risks, and impose financial penalties. The obligation applies whether you build and release a model or deploy one that someone else built.
Who it binds
GPAI model providers (anyone releasing a general-purpose AI model). Deployers of any AI system that interacts directly with people or generates synthetic content (text, image, audio, video, code). In practice: model companies, API platforms, chatbot operators, content-generation tool makers, and organizations integrating such tools into customer-facing products.
How to check whether you comply
Audit your product interfaces: does every interaction with an AI system clearly signal that to the user? Check in-app labels, tooltips, and consent flows.
Review content outputs: are synthetic images, videos, audio, and text marked before delivery to users? Test the labeling on outputs at scale.
Document your deepfake disclosure process: how do you tell users when content is manipulated or synthetic? Verify the mechanism is live in all user-facing channels.
If you provide a GPAI model: prepare documentation of model capabilities, training data lineage, and known risks. The AI Office may request this.
What teams get wrong
This deadline has passed as of August 2, 2026. If you have not implemented transparency labeling and disclosure, you are already non-compliant and subject to enforcement. The penalty regime is now active, meaning fines can be issued. Compliance is not optional or phased; the obligation went into force on the stated date.
AI Act — Article 50 marking, systems already on the market — EU, 2026-12-02
AI Act — Article 111(3), legacy GPAI models — EU, 2027-08-02
Source
This page is part of the AI Compliance Deadline Radar, which tracks every dated AI obligation across the EU, UK, and US. Updates go out in The AI Reg Brief, twice a month.
Not legal advice. This summarizes a public source so you can act on it; verify against the primary text before you rely on it.