Tag
Research finds that AI text watermarking alters language model responses to harmful prompts, potentially increasing vulnerability to adversarial attacks and affecting agent behavior.