{
  "id": 1716247,
  "title": "How to avoid Claude watermarking your content",
  "url": "https://urgent.news/2026/08/18/how-to-avoid-claude-watermarking-your-content",
  "topic": "ai",
  "section": "AI",
  "published": "2026-08-18T13:30:27.000Z",
  "source": {
    "name": "ZDNet",
    "slug": "zdnet",
    "url": "https://www.zdnet.com/article/how-to-avoid-claude-watermarking-your-content/"
  },
  "original_language": "en",
  "account": "Anthropic has announced it will start watermarking text output from several of its LLMs, such as Claude, to comply with EU AI Act regulations. However, this has led to public outrage, prompting researchers to search for ways to avoid the watermarks. Writing it yourself is the only solution, but many are looking for alternative fixes. The watermarks work by attaching to text based on specific word choices, making it possible to detect Claude-generated content. This random word selection process is changed slightly by the watermarking key. Watermarking may seem like a minor issue, but it raises questions about the role of AI tools and the importance of content provenance. Users need to be more aware of the limitations and expectations of AI tools, and focus on writing themselves instead of relying on AI.",
  "summary": "The answer is where you least expect it.",
  "key_points": [
    "Anthropic will watermark Claude's LLM output to comply with EU AI Act.",
    "Watermarks attach to text based on specific word choices.",
    "Researchers seek alternative methods to avoid watermarks."
  ],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}