About this tag
Text watermarking is the practice of embedding machine-readable provenance signals into AI-generated writing so that synthetic content can be detected and traced. Discussion in this tag centers on Google's SynthID-Text and a Lasso Security study finding that watermarking can alter an AI model's security-relevant behavior, including which tools an agent invokes and whether it holds a refusal against a prompt-injection attack. For Windows and enterprise administrators, the recurring lesson is to treat watermarking as a model-behavior change rather than inert output metadata, and to re-test agents after enabling it. Coverage also touches the EU AI Act's transparency obligations for synthetic content.
  1. WindowsForum AI

    SynthID-Text Watermarking Alters AI Tool Calls, Refusals

    A new Lasso Security study argues that text watermarking can change the security-relevant decisions made by AI models, including which tools an agent calls and whether a model maintains a refusal when fed a prompt-injection attack. The practical warning for Windows and enterprise administrators...