Post

Stephan Lewandowsky
@lewan.uk

Very pleased to share a new policy brief by CAISA: Governing AI-generated content: Labelling for transparency when AI is the norm. This was a genuine privilege to be part of the team of 9 leading European scholars caisa.dk/forskning/t... 🧵 1/10

9 reposts29 likes

Stephan Lewandowsky
@lewan.uk

The context: AI-written content may have surpassed human-created content online in 2025. Citizens can't reliably tell the difference between AI and human content. So what do we do? The EU AI Act's transparency provisions take effect on 2 August 2026. 2/10

Stephan Lewandowsky
@lewan.uk

The EU AI Act (Article 50) requires that AI-generated content be marked in machine-readable formats and disclosed to users. A Code of Practice on this is being finalised right now — its final version drops in June. 3/10

Stephan Lewandowsky
@lewan.uk

But here's the problem: labelling alone won't cut it. Research by my team shows that even when people are explicitly warned a video is a deepfake, they are still influenced by it. Transparency is a normative good, but not a shield. 4/10

Stephan Lewandowsky
@lewan.uk

Helberger asks the key question: are users informed or empowered? A generic "AI" tag doesn't say what was manipulated, why, or whether the content is trustworthy. It can even backfire, generating mistrust of accurate AI-assisted content. 5/10