SynthID-Text

SynthID-Text is DeepMind's watermarking technology designed to identify AI-generated text by embedding imperceptible statistical signatures during the generation process. This tool helps organizations detect synthetic content in their systems and combat misinformation at scale. For AI governance, SynthID-Text enables provenance tracking and authenticity verification, which are critical for enterprises managing content authenticity risks and ensuring transparency about AI-generated materials in compliance frameworks.

1 item

ResearchGlobal2026-09-17

SynthID-Text Watermarking Weakens Safety Guardrails, Lasso Security Finds

Lasso Security researcher Andrea Siposova found that SynthID-Text watermarking alters how LLMs respond to harmful prompts, including bypassing safety refusals. The effect, which Siposova calls 'sampling drift,' extends into agentic pipelines by influencing which tools agents invoke. Anthropic has committed to deploying SynthID-Text in future Claude models, partly in response to EU AI Act provenance requirements.

watermarking SynthID-Text safety guardrails agentic AI Anthropic Google DeepMind EU AI Act red teaming