AI transparency obligation
The duty to tell people when they are dealing with AI or seeing certain AI-made content, instead of letting the machine pass as human or real.
What it is
An AI transparency obligation is a duty to make the machine visible. Under Article 50 of the EU AI Act, examples include telling people when they are interacting with an AI system, marking certain synthetic outputs in a machine-readable format, and disclosing specified deepfake content. The exact duty depends on whether you provide or deploy the system and what the system does.
Reach for a disclosure at the point where the person encounters the AI, not three links deep in a privacy policy. A chatbot should identify itself before the conversation starts; a synthetic image or recording needs a visible label where required, plus durable provenance metadata such as C2PA Content Credentials where the provider controls the file.
Gotcha: one generic 'may use AI' sentence does not cover every use. Generating a support reply, ranking a loan application, recognizing emotion, and publishing a deepfake can trigger different rules and different owners. Inventory each AI touchpoint and assign its specific notice, timing, format, and evidence.
Ask AI for it
Implement an AI transparency layer for the product described below against Article 50 of the EU AI Act. Inventory every user-facing AI touchpoint and classify the product as provider or deployer for each. Before the first chatbot interaction, show a persistent 'AI assistant' label and a plain-language notice of its limitations. For generated images, audio, and video, attach C2PA Content Credentials when the file pipeline supports them and add the visible disclosure Article 50(4) requires for deepfake content. Record the notice version, locale, placement, and timestamp in an audit log. Do not hide the disclosure in the terms or privacy policy. Mark unresolved role, exemption, and jurisdiction questions as bracketed TODOs for counsel.