Deepfake disclosure: the transparency rule

The AI Act's rule for deepfakes is straightforward: if content is artificially generated or manipulated to appear authentic, you must disclose that fact. This applies to images, audio, and video that could deceive a person into thinking they are real. The key contrast is with minor edits like colour correction or harmless filters, which do not count as deepfakes. The regulation aims to protect people from being misled, especially in contexts like news, politics, or personal reputation.

For the exam, remember that disclosure is the default, but there are limited exceptions. For example, deepfakes used for satire, artistic expression, or legitimate law enforcement purposes may be exempt. A quick elimination trick: if a question lists a deepfake without any context of exception, choose the answer requiring disclosure. Another hint: look for phrases like 'for entertainment only' or 'clearly labelled as parody'—these signal a possible exception. Avoid assuming all deepfakes are banned; the rule is about transparency, not prohibition.

To test yourself, ask: 'Would a reasonable person be fooled by this content without a label?' If yes, disclosure is needed. A memory aid: think of a deepfake as a 'digital mask'—you must always say it's a mask unless it's obviously a costume party (satire or art). This keeps the focus on the transparency principle, not on blanket bans.