Anthropic's plan to watermark Claude-generated text is prompting questions and unease among some figures in tech.
The AI lab announced this week that some Claude models will embed an "imperceptible watermark" directly into their text output, which will travel with the content when it is copied and pasted and may survive some editing, helping distinguish AI-generated material from human writing.
Anthropic has not yet publicly shared full technical details on exactly how the system works, which has led some techies to raise concerns about copyright, output quality, and how lightly AI-assisted work could be judged.
Who can see the watermark? #
One concern, raised by tech investor Bill Gurley, is that only Anthropic would be able to identify the watermark, which he said would make the company "judge, jury, and prosecutor."
An Anthropic spokesperson told Business Insider that it plans to roll out a free application programming interface, or API, that will allow users and third parties to check text for Claude's watermark themselves.
What about content that's been lightly edited by AI? #
Some developers also questioned whether watermarking could affect Claude's answers if it used a technique that combined specific words that would flag it as AI-generated. The Anthropic spokesperson said the watermark "doesn't change the meaning, quality, or readability of Claude's responses."
Simon Smith, who leads generative AI at digital health agency Klick, said in Tuesday X post that one of his concerns was that lightly edited AI work, such as a grammar check run through Claude, could be flagged as AI-authored.
Anthropic said a watermark shows that Claude processed text, not necessarily that it wrote it, and that the mark can remain after Claude has proofread, translated, or summarized content.
What does this mean for copyright? #
Steven Sinofsky, a former Microsoft executive, raised concerns about digital privacy. "The real issue is data retention and your right to private thoughts free of a digital trail," he wrote in an X post on Tuesday.
Meanwhile, software engineer trainer and coach John Crickett questioned in an X post on Tuesday whether a watermark on AI-generated code could complicate copyright claims because authors wouldn't be able to show sufficient human input.
Anthropic told Business Insider that it is adding the watermark — which has applied to Claude-generated text globally on supported models since August 2 — in order to comply with the European Union's AI Act.
Anthropic isn't the only AI lab to use watermarks: Google uses its SynthID technology to watermark AI-generated content, and OpenAI uses SynthID for supported images and audio.
Elon Musk's social media platform X also adds a "Made with AI" tag on content it determines to be AI-generated or manipulated. The tag appeared on the resignation post of outgoing White House Press Secretary Karoline Leavitt on Wednesday, but later disappeared.
Claude's watermark could have benefits #
Not everyone sees watermarking as a mistake.
Software developer Donn Felker said on X on Tuesday that it could help prevent AI systems from being trained on a growing volume of AI-generated material, creating what he called a "snake eating itself" problem, in which AI systems degrade as they are trained on their own outputs.
Aadit Sheth, cofounder of executive communications firm The Narrative Company, also welcomed the added transparency. He said on X on Tuesday that the watermark would help others identify AI-generated writing, adding that audiences should be able to tell whether the words they read reflect a person's own thinking.
One of the more obvious impacts of Anthropic's watermark is that it would expose people for using AI for everyday tasks like drafting emails or writing LinkedIn posts.
"How much do people care if their AI use is outed?" wrote Business Insider's Katie Notopoulos. "If we look at the most common uses of what people actually use AI to write, I suspect the answer is less than we might imagine."