Anthropic's SynthID watermarks — why users are canceling subscriptions
anthropic detailed how its syntheid-text watermarking works for claude to comply with eu transparency codes, prompting user cancellations.
anthropic confirmed it is using google deepmind's syntheid-text approach to watermark claude outputs, creating a statistically detectable pattern in low-stakes word choices to comply with the eu ai act's transparency code. the company claims the watermark survives light editing but may vanish under complete rewrites, though it will be negligible in functional code segments to preserve validity. anthropic plans to release a detection api and notes that other major developers have signed the same code of practice, standardizing watermarking across the industry despite user backlash over privacy and detection concerns.