Anthropic’s new watermark for AI-generated text struck a nerve for some Claude users — but did Europe’s laws require it to go as far as it did?
A Business Insider review of the law, supported by interviews with AI and policy experts, suggests Anthropic’s approach may go beyond what the EU requires.
Under the European Union’s Article 50(2), AI companies do not have to mark AI outputs when the technology is used for standard editing or does not substantially change a user’s original words or meaning.
Anthropic’s watermark, which is designed to remain detectable when text is copied and pasted, can appear in translations, summaries, and more extensively edited text.
“Basically, Article 50 called for a scalpel to catch deepfakes and mass synthetic generation, but Anthropic is using a sledgehammer that hits simple proofreading and translation,” Caroline De Cock, senior fellow at the Lisbon Council think tank and head of research at the Information Labs, told Business Insider.
Anthropic has said it is introducing the watermark to comply with the EU’s AI Act and that it’s rolling it out globally because it does not have a way to apply it on a per-region basis.
While some have welcomed the transparency that AI watermarks can bring, Anthropic’s announcement also triggered a wave of backlash. Some Claude users said they canceled their subscriptions, and developers have created new tools to remove the watermarks.
A recurring concern is that the watermark could be interpreted as indicating that Claude wrote the text, even if only used for editing or proofreading.
This is due to the technical approach behind Anthropic’s watermark: it guides Claude’s word choices to create a detectable pattern, indicating that the chatbot was likely involved in writing the text.
Lighter editing with Claude is less likely to trigger the watermark, an Anthropic spokesperson told Business Insider, because “there’s very little (if anything) for the watermark to attach to.”
Milos Rusic, the cofounder and CEO of enterprise AI company Deepset, told Business Insider that Anthropic “is taking a somewhat broader interpretation” of the EU law. That’s because there can be a “provenance signal” attached to Claude’s word choices, even when it “did not originate the underlying idea or document,” he said.
So, why might Anthropic choose this approach?
It’s in line with Anthropic’s “brand identity as a safety-first AI developer,” De Cock said, adding that it could also reduce its regulatory risks by “shifting the burden of filtering and interpreting those labels entirely onto downstream users and external platforms.”
An Anthropic spokesperson told Business Insider that it would “continue to evaluate different approaches.”
A wave of AI watermarks
The EU’s watermarking laws apply to new models launched in Europe on or after August 2. Older models have a grace period until December.
That means other tech companies operating on the continent will have to roll out their own machine-readable ways to detect AI-generated content in the coming months.
Some already have them.
Google said it uses its SynthID technology to invisibly watermark AI-generated text, images, audio, and video from its models worldwide.
OpenAI referred Business Insider to an FAQ saying it plans to expand its labels to text, but did not specify a date.
Meanwhile, Meta directed Business Insider to a July blog post about its work on practical, interoperable tools to identify AI-generated content. Meta has so far focused on images and has not provided details on a text-watermarking plan.
xAI did not respond to a request for comment from Business Insider. The company tags some work as AI-generated, including, recently, White House press secretary Karoline Leavitt’s departure message on X. The tag later disappeared.
The coming wave of watermarks comes at a time when many companies have been encouraging their employees to use AI. Now, some people are worried about it being outed and their authorship of work being questioned.
“These systems can be useful evidence,” Rusic said of watermarks, “but I would be very cautious about turning them into an automated judgment about authorship or misconduct.”
Read the full article here
