Anthropic quietly adds invisible watermarks to Claude's AI-generated text

On August 14, 2026, Anthropic revealed that Claude's responses now carry hidden watermarks, based on Google DeepMind's SynthID-Text method published in Nature in 2024. The watermark slightly shifts which words the AI picks, leaving a pattern people cannot see but tools with the right key can detect. This is among the first real-world uses of AI text watermarking in a major chatbot.