Anthropic announced it would watermark AI-generated text to distinguish it from human content. This involves subtly altering word selection randomness. John Gruber strongly opposed this change, comparing it to an unwelcome alteration of a beloved item. A significant portion of this opposition likely stems from Anthropic ostensibly complying with an EU mandate. However, it's suggested Anthropic might be watermarking to prevent retraining its own output, thus avoiding model degradation. The company might also want to identify and exploit competitors' watermarks while keeping their own secret. The author's disapproval of Anthropic's plan is independent of any EU regulation. They believe the EU regulation itself is flawed and already outdated. The author thinks the EU regulation is a distraction from the core issue of detectable word choice watermarking. The author suspects Anthropic and Google are primarily motivated by self-serving reasons related to model training. Blaming the EU regulation is seen as scapegoating by Anthropic. John Moltz also offered a notable critique of MacStories' return to Twitter/X.