Study: Watermarking AI Text May Increase Hallucinations

An AI Insider analysis suggests that watermarking AI-generated text can raise the likelihood of hallucinations, as embedding the mark interferes with token generation. The trade-off between provenance and accuracy remains unresolved.

• Watermarking AI text may trigger more hallucinations
• The mark is embedded during token generation
• Trade-off between provenance and accuracy is unresolved