Anthropic has released an FAQ explaining its watermarking implementation for EU AI Act compliance. The company clarifies that watermarking adds no visible changes to outputs, requires no extra tokens, and introduces no hidden characters. Other major model developers have signed the same Code of Practice.