The Problem
Why deepfakes represent a systemic threat
Concrete risks
Political disinformation
Fabricated videos of political figures can influence elections, cause panic or trigger diplomatic crises. Deepfake-based disinformation campaigns have multiplied during every major election in recent years.
Identity theft
Ordinary people find their face and voice used without consent for fraud, harassment or non-consensual pornography. Generative AI has made these attacks accessible to anyone.
Financial fraud
Audio-visual deepfakes of company executives are used to authorise fraudulent wire transfers. Several documented cases have involved losses of millions of euros.
What is happening today
Several deepfake detection tools exist, but they regularly fail on compressed, edited or next-generation content. The race between generation and detection is structurally unbalanced: generative models advance faster than detectors.
The platform problem
When a video is published on a platform, it is compressed and re-encoded, often stripping the original metadata — including any provenance markers. If the law does not require platforms to preserve and display this information, any labelling done at the source becomes useless.
Why detection alone is not enough
Waiting for an automated tool to detect deepfakes always means intervening after the fact, on content that has already been shared. Mandatory labelling at the source is the only preventive and robust long-term approach.
Sources: Europol, AFP Fact Check, MIT Media Lab, Centre for Emerging Technology Policy.