The rationale for a law requiring AI text watermarking could have been applied to calculators, but thankfully, human and calculator output are currently treated similarly. Similarly, spellcheckers are not scrutinized for their assistance. Singling out AI assistance only when it can compose entire sentences is not a principled distinction but a misplaced value on difficulty. Anthropic's broad, model-level watermarking implementation exceeds the law's minimum requirements. This approach disregards distinctions the law intended to maintain. The resulting watermark is sufficiently wide to affect innocent and helpful uses. However, it can be easily removed by determined individuals through significant rephrasing. This risks unfairly targeting everyday and assistive users while being least effective against deliberate cheating. James Padolsey, creator of Declaude, highlights this issue, with his app capable of removing Claude's distinctive style and potentially these new watermarks. Padolsey's work demonstrates the poorly conceived nature of this EU regulation, as Anthropic and Google Gemini adopt similar watermarking techniques.