NEW YORK, NEW YORK - FEBRUARY 16: In this illustration, the Claude AI app is seen in the app store on a phone on February 16, 2026 in New York City. According to reports from the Wall Street Journal, the Defense Department used Anthropic's Claude Ai, via its Palantir contract, to help with the attack on Venezuela and capture former President Nicolás Maduro. (Photo illustration by Michael M. Santiago/Getty Images)

Getty Images

Like many professionals, I use Claude and various other AIs to proofread what I write. Sometimes I ask for suggestions for specific sentences to be rewritten for clarity or conciseness. With the recent announcement by Anthropic that watermarks will be added to all text, I, like many people, have no idea if this means my work is now marked as AI-generated. While the action has generated valid discourse on whether watermarking is a good idea, or whether the specific technical method used by Anthropic is the right one, that is not the focus here. Rather, I see this as a trend in development. Regardless of how these debates resolve and when, they are likely to lead to more ambiguity in the near future for businesses. How should businesses navigate AI usage and protect ROI, while governments, providers, and technologists try to connect the legal and technical landscapes?

First, What Happened?

Earlier this month, the EU Code of Practice on Transparency of AI-Generated Content took effect. Around 200 companies signed on to this agreement in the abstract (including Meta, Microsoft, OpenAI and Anthropic). Anthropic took the first visible step in operationalizing this practice by creating a global opt-out watermark applied at model level across all Claude surfaces.

Since then, there has been a wide range of reactions, from an immediate proliferation of tools claiming to remove watermarking to Anthropic seeking to address everyday users’ fears about what this means.

Watermarking Demystified

Watermarking can sound like technical magic. It isn't. Here's a simplified picture of the idea, not a technical account of what Anthropic actually built. There are many technologies that Anthropic may have used or developed from, ranging from incorporating special words, hidden text patterns, special characters, or conducting analyses of patterns found in AI-generated content that are not found in normally human-generated text. A very simple example (very likely NOT what Anthropic is using)

• Say the AI generated a text “My cat is fluffier than my dog.”
• Within the spaces, the AI can insert characters that are not displayable commonly on computer screens but exist and will persist across copies. For example, the text may actually say “My<special character>cat is fluffier than my dog”.

This is a very simple example, and one easily debunked by running the text through a reader that lists all the characters. Production watermarking methods are far more sophisticated.

MORE FOR YOU

Should My Business Know The Technology?

From a business perspective, which technology Anthropic is currently using is not the key.

• Anthropic (or other AI companies) can change watermarking technologies at will.
• Watermarking is an active area of research, so even if a given watermark is detectable and removable by third-party tools, that does not mean future text will work the same way.
• Given that this development was in response to a legal practice coming into effect, one can assume further evolution will occur as all parties understand what works and does not work for the legal mandate.

So, in short, unless your company is in the business of watermark generation or removal, or you need to watermark your own products, the specific technology in use at any time is not the key element.

Precedent, Not Verdict.

So what is the key? The key is, regardless of what anyone thinks of watermarking, one company has now set the de facto implementation standard others get measured against, whether they intended to follow it or not. Anthropic’s move is likely the first of many in this space, with watermarks being constantly generated, possibly detected and removed, and the cycle repeating with ever-advanced technologies.

What Can Businesses Do?

The first step is to recognize the ambiguity this scenario creates, and accept that the ambiguity is here to stay for the foreseeable future/

• Any AI tool that your business uses can generate hidden markings in the output, recording the use of the tool. As these watermarks are designed to do, the mark will persist across copies of the content, and may or may not change depending on how much the content is edited.
• Governments and other qualified parties may have received a watermark detector. Given the early nature of this space, the efficacy of detectors also remains to be seen.

The second step is to appreciate the intellectual property questions this raises. If one of your employees wrote a document themselves, and then asked AI to edit it, would the edited document contain the watermark? Would it be clear how much of the work was done by the employee and how much by the AI? Would this affect whether your company owns the intellectual property of the document? These are questions that will only be answered with time.

The ambiguity isn't only in how businesses interpret the watermark. It may be built into the watermark itself. Business Insider reported that Anthropic's approach could extend beyond what EU law requires: Article 50(2) exempts standard editing that doesn't substantially change a user's original text, yet Anthropic's watermark can persist through translation, summarization, and other forms of editing. Google, Meta, and OpenAI have each weighed in with their own responses on where they stand on AI watermarking, and Business Insider lays out the details of each company's approach for those who want to compare.

Provenance is the core response.

In the presence of such ambiguities, businesses should establish as much provenance tracking as possible.

• Establish provenance for anything that touches AI, including the organizing idea.
• Brief employees on what is and isn’t yet understood. Acknowledge the ambiguity rather than pretend certainty.
• Set internal guidelines for what can/can’t be AI-generated and how AI use is documented. Make these guidelines independent of what any provider does at any time. They should be appropriate to your business ROI, not the AI or AIs used in your workflows.
• Name one owner, ideally in your C-suite, accountable for AI's IP implications.

Corporate Taste

In a prior article, I covered why Corporate Taste is the new business moat. This is particularly true as AI generates ambiguity in business operations, whether due to discontinuities created by access suspension to models undergoing export control restrictions, or models adding watermarks that create questions of ownership. These kinds of changes are likely to grow rather than ebb over the next decade as AI continues to revolutionize the way companies operate. This is why Corporate Taste is increasingly important as an asset to identify, grow, secure, and protect. Close. ROI is the non-negotiable goal. Ambiguity is the non-negotiable operating reality. Corporate Taste is what lets you hold both at once

Takeaway

To protect business ROI, it is not just enough to be aware of legal precedents such as the EU Act, or its operational precedents like the watermarking actions by Anthropic. Business practices to enhance governance and clarify ownership of these priorities will help your business protect its Corporate Taste and, consequently, its ROI.