{
  "id": 1492350,
  "title": "AI-generated text should be detectable, but Apple needs to avoid Anthropic’s huge error",
  "url": "https://urgent.news/2026/08/17/ai-generated-text-should-be-detectable-but-apple-needs-to-avoid",
  "topic": "ai",
  "section": "AI",
  "published": "2026-08-17T13:30:47.000Z",
  "source": {
    "name": "9to5Mac",
    "slug": "9to5mac",
    "url": "https://9to5mac.com/2026/08/17/ai-generated-text-should-be-detectable-but-apple-needs-to-avoid-anthropics-huge-error/"
  },
  "original_language": "en",
  "account": "Writing is my profession, and I wholeheartedly support the idea of making AI-generated text identifiable, marked with an \"AI content\" label. However, a recent announcement by Anthropic has raised concerns: they are complying with an EU initiative to watercolor AI-generated text using Claude, but in an unusual way. Apple is also preparing its response to the issue of AI-generated imagery through a feature called Apple Reference Image, which may be extended to Siri AI tools as well. The most daring requirement from the EU initiative mandates both text and image outputs to be digitally watermarked. This can be accomplished using invisible characters, like the one I used to replace the space between the words 'invisible' and 'characters,' which remains intact even after copying and pasting. Optical character recognition could easily defeat this, however. I had proposed a different approach, including using specific language patterns in Claude output to detect even OCR text. But now it seems Anthropic has adopted this method, prompting me to comment. Embedding language patterns in entirely AI-generated text is not an issue for me, as long as it doesn't affect human-written work, where AI tools are only used for proofreading or editing. The problem arises when Claude embeds these hidden watermarks even when asked to identify and correct grammatical errors in human-written content. This issue is similar to Apple's Siri AI, which, while testing, proved too aggressive in its proofreading suggestions. I would prefer that such a tool remains focused solely on detecting grammatical errors, offering minimal correction. In the case of Apple, I believe it would be inappropriate for an AI to deliberately alter a human author's words for the purpose of watermarking. Similarly, marking AI-assisted edits as AI content if they only perform sub-editing would be incorrect. Instead, a 'Proofread by AI' label would be more fitting. While I don't claim to have all the answers, the growing reliance on AI chatbots for writing and editing presents complex questions. When a human's work is edited by AI, when should it be recognized as a hybrid creation of the author and the AI system? What I am certain about is that intentionally modifying a human author's chosen wording, based solely on AI's proofreading assistance, is not the right approach. While Apple will have to address this issue eventually, I hope it will adopt a more sensible solution than Anthropic's. What are your thoughts on this matter? Please share your opinions in the comments. Check out 9to5Mac on YouTube for more Apple news.",
  "summary": "As someone who writes for a living, you would correctly guess that I’m wholeheartedly in favour of allowing AI -generated text to be detectable and marked as such. There’s just a ridiculous amount of AI slop out there, and an “AI content” label means I don’t need to waste my time reading any of it. However, Anthropic has just announced that it’s complying with an EU initiative to have Claude…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 2,
    "also_reported_by": [
      {
        "outlet": "Guardian Technology",
        "title": "Claude to start watermarking AI-generated text – but will it make quality worse?",
        "url": "https://urgent.news/2026/08/17/claude-to-start-watermarking-ai-generated-text-but-will-it-make",
        "published": "2026-08-17T16:52:04.000Z"
      }
    ]
  },
  "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."
}