As participants in the Code of Practice drafting process, we provided feedback on the First Draft Code of Practice on Transparency of AI-Generated Content, covering both Working Group 1 (marking and detection) and Working Group 2 (disclosure requirements).
Overall we felt that the first draft was a good one and the writing process is off to a good start. But at the same time several points need to be improved. Our key recommendations included:
- Need to engage with feasibility constraints: Watermarking cannot be robust for very small content (short texts, small images). Requirements should acknowledge these limits and use size-dependent metrics rather than absolute thresholds.
- Proportionality: Requirements should distinguish between system providers and model providers, as Article 50(2) obligations apply only to the former. SMEs and open source projects require proportionate pathways.
- Disclosure mechanisms: Provide specific guidance on icon placement across modalities whilst ensuring proportionate disclosure for creative works that preserves artistic expression.