Compliance checklist

Article 50 of the AI Act: the obligations of 2 August 2026.

Attention goes to high-risk systems and prohibited practices. Yet it is Article 50, the transparency obligations, that reaches the largest number of organisations: it applies to any generative AI system, whatever its risk level, and open-source systems are not exempted.

Four situations trigger an obligation

Direct interaction with a person

A system designed to interact with natural persons must inform them that they are dealing with an AI, unless obvious from context. The draft guidelines specify this also covers agentic systems.

Production of synthetic content

Providers of systems generating audio, image, video or text must ensure outputs are marked in a machine-readable format and detectable as artificially generated.

Emotion recognition, biometric categorisation

Deployers must inform exposed persons of the operation of the system.

Deepfakes and texts of public interest

Deployers producing a deepfake must disclose that the content was artificially generated or manipulated; an AI-generated text published to inform the public must be disclosed too.

The eight checks to run

The checklist details each check: what it covers, who owns it in the organisation (general management, legal, communication, IT), the typical mistake seen in practice, and a model notice or clause to adapt. Including the most neglected check (your own conversational interfaces, already in production) and the most sensitive one (your published texts on matters of public interest). Document in French.

Map the AI systems in use, including open source
Qualify your role, provider or deployer, for each system
Check the disclosure of conversational interfaces
Address machine-readable marking of produced content
Plan the visible disclosure of published deepfakes
Review your published texts on matters of public interest
Address emotion recognition and biometric categorisation
Contract with your suppliers, and keep a trace of your analyses

The real timeline

Application of the transparency obligations: 2 August 2026. That is settled. Upstream, the Commission published draft guidelines (8 May 2026) then a code of practice (10 June 2026). A deferral to 2 December 2026 was envisaged for the machine-readable marking obligation only, for systems already on the market; the checklist explains why pausing your preparation on that basis would be a disproportionate risk.

For general management, legal, communication, compliance and IT. Written in July 2026. AI law moves fast; check the state of the text before any decision.

Flagship track

The checklist sets the floor. Our programmes take it from there.

Epistemic Security and Information Resilience

Article 50 mandates a minimum of transparency. But deciding in an environment where information is manipulated (deepfakes, synthetic content, coordinated campaigns) takes more than compliance. That is the purpose of our executive track, grounded in doctoral research.

Explore the track See the full catalogue

Our programmes

The programmes directly tied to the AI Act

Four tracks for the teams concerned by Article 50: governance and legal framework, deepfake detection, generative AI for executives, and executive information security. All OPCO-eligible.