https://www.getreadyforagents.com/news/synthid-text-watermarking-adversarial-prompt-vulnerability/
SynthID-Text watermarking alters LLM responses to adversarial prompts, increasing safety guardrail...
Sep 19, 2026 - According to Ars Technica research, Anthropic's planned use of Google's SynthID-Text watermarking in future Claude models can change tool invocation and model a
synthid text watermarkingaltersllmresponsesadversarial