3 arguments for and against AI watermarks

Anthropic is watermarking AI-generated text. The news divided techies. OJO Images/Getty Images Anthropic's invisible Claude watermarks are dividing tech over transparency and privacy concerns. Supporters say watermarks could expose undisclosed AI use and curb AI training on AI slop. Watermark critics say the tech can flag lightly edited human work and create compliance risks. Anthropic's move to place invisible watermarks within Claude - generated text is dividing the tech community. Claude models launched on or after August 2 will have an "imperceptible watermark" embedded in their text output, and Anthropic says it also plans to add support for older models. The AI lab says the mark travels with copied-and-pasted text and that it plans to release a detection tool soon. That has set off a broader debate : Should people be able to tell when AI was used, or should users be able to work with AI without leaving a permanent mark? Here are three arguments for and against it. For: Watermarks mean greater transparency From social media posts to job applications to schoolwork, it's hard to know for sure what's AI-generated versus human-made. Supporters say readers deserve to know when the words they are reading were generated by a chatbot. Aadit Sheth, the cofounder of executive communications firm The Narrative Company, said in an X post that people should be able to tell whether a piece of writing reflects someone's own thinking. "When you use AI, you know yourself that you are not being true to yourself," Sheth wrote. "It's another system speaking." Against: The tech isn't perfect and could expose people in high-stakes settings Anthropic is also quick to warn that its watermarking text has limits, saying it can only determine that Claude was "likely involved with the content at some point." One concern is that nuance could be lost, and people will just assume any text that tests positive for the watermark was fully generated with AI. Some users fear that a watermark could create problems with clients, bosses, schools, or contracts when they've used AI to assist their work. Freelance developer Vladislav Rajtmajer told Business Insider that he was worried that a Claude marker in code sent to a client could raise questions about authorship. Anthropic has said the watermark will have a "negligible effect" on code. Arturo Villarroya, who works in public policy, said he feared trouble in professional and academic settings with strict AI rules. Anthropic, for its part, says its watermarking can't be traced back to a certain organization or individual users. The company says it's adding watermarking to meet its commitments to the European Union's AI Act. For: They could help professors catch cheating students Within academia, AI cheating has grown from a rumble into a roar. Students have told Business Insider that they feel guilty about their AI usage for schoolwork, with some fearing over-reliance or skill atrophy. Universities are implementing stricter policies, such as in-person exams and laptop-free classrooms . Still, many educators cannot keep up with all the complex ways students skirt the rules with AI. The watermark could provide another tool. "That will be awesome in universities," wrote Gaetan Semet , a software developer for Renault Group. "They will all get caught." Against: The editing question One of the loudest criticisms is that Anthropic may watermark work that had minimal input from Claude. Anthropic says a mark can remain after someone uses Claude to proofread, translate, or summarize text they wrote themselves. Analyst Ben Thompson, writer of the Stratechery newsletter, said that forcing a writing tool to claim credit is like asking a ballpoint pen to advertise itself as the author. "It hardly seems fair that anyone who wants to fix their grammar runs the risk of having their original content labeled as AI," he wrote. Anthropic responded to the proofreading question in a blog post. The watermark only applies to words chosen by Claude, it wrote. "Because nearly all the words are the person's, there's very little (if anything) for the watermark to attach to." For: Watermarks could help AI companies avoid training on AI slop There is another argument for labeling AI output: tech companies may not want to train their next models on a flood of text made by earlier models. Software developer Donn Felker called it a "snake eating itself" problem. If chatbots keep ingesting their own output, the quality of future systems could suffer. "No one seems to be talking about this, but at some point, we could have a real problem where we can no longer improve AI because everything is already AI-generated," Brett Levenson, CEO of AI governance company Moonbounce and a former Meta trust-and-safety lead, told Business Insider. "I think Anthropic is simply trying to delay this moment — or at least be aware of it when it arrives," he added. Against: They could make copyright claims more difficult It may be trickier to prove that you own your work. Copyright has long been a big question in the age of AI: If I use AI to generate a work, is it actually mine? The watermark could blow that question open, as AI use becomes more visible. Software engineer trainer and coach John Crickett applied the theory to AI-generated code. "If it's watermarked as AI generated, how can the author claim it had enough human input to claim copyright?" he asked. In response to the question of whether AI watermarks "change who owns a given output, or who is legally responsible for it," Anthropic wrote in its blog post that the answer is "No." "A watermark only helps test whether Claude might have produced or processed the content," Anthropic wrote. "It doesn't say anything about ownership or authorship, and doesn't change a user's rights under our terms." Are you for or against AI watermarking? Tell us your view in the comments below: Read the original article on Business Insider