AI Watermarks: Does Labeling Still Make Sense?
In one line: Skepticism is growing over whether watermarking and labeling AI-generated content is actually effective.
Key points
- As AI images, video, and text become harder to tell apart from human-made work, the debate over the value of "made by AI" markers has reopened.
- Watermarks can be stripped or degraded through cropping, re-encoding, and editing, exposing technical limits.
- Even as labeling mandates advance, users are reportedly left to judge the trustworthiness of unlabeled content on their own.
Why it matters
Watermarks are a linchpin that regulations such as the EU AI Act lean on. But if markers are easily erased and unlabeled content proliferates, labels risk burdening only honest creators while bad actors slip through — an asymmetry that undermines the goal. Verifying that disclosure methods actually work is emerging as the next challenge in regulatory design.