From emerging regulation to deal mechanics, we write about the questions founders and investors actually ask — practical analysis you can put to work.
Synthetic content is audio, image, video or text generated or significantly manipulated by AI. Under AI Act Article 50 it must carry machine-readable marking from 2 August 2026, with visible labels for deepfakes and public-interest text.
Under the EU AI Act, a deepfake is AI-generated or manipulated image, audio or video content resembling existing persons, objects, places or events that would falsely appear authentic — triggering a visible disclosure duty from 2 August 2026.
AI watermarking embeds detectable signals into AI-generated audio, image, video or text so machines can identify the content as synthetic — one of the techniques providers can use to meet the AI Act’s Article 50 marking duty.
Content Credentials are cryptographically signed provenance metadata based on the C2PA standard, recording how a media file was created and edited — the emerging de facto answer to the AI Act’s machine-readable marking duty.
The TDM exceptions (DSM Directive Arts. 3–4) permit mining lawfully accessed works — but commercial mining is subject to rights-holder opt-outs, which AI Act Article 53 obliges GPAI providers to respect through a copyright policy.
The training content summary is the public document GPAI providers must publish on the AI Office template, describing the data used to train the model — including for fine-tuned modifications, covering the modification’s data.
Article 50 requires disclosing AI interaction to users, machine-readable marking of AI-generated content, visible deepfake labelling and emotion-recognition notices — applicable from 2 August 2026 and not postponed by the omnibus.
A provider develops an AI system or GPAI model (or has one developed) and places it on the market or puts it into service under its own name or trademark — the role carrying the heaviest AI Act obligations.
A deployer is any natural or legal person using an AI system under its authority in the course of a professional activity — the AI Act role that carries use-side obligations such as human oversight, input control and, for some systems, a FRIA.
A GPAI model is an AI model trained on broad data at scale that can competently perform a wide range of tasks and be integrated into many downstream systems — the EU AI Act’s category for foundation models, regulated since 2 August 2025.
