The Watermark Changes Nothing: AI Is Still a Tool,Not the Author Revisiting Generative AI: A Tool, Not a Shameful Shortcut in the age of AI provenance

In January 2025, I published an article titled Generative AI: A Tool, Not a Shameful Shortcut.

My argument was straightforward: using generative AI to help articulate an idea should not automatically diminish the value of that idea. Writers have always worked with tools, editors, researchers, dictionaries, search engines, spell-checkers, designers and collaborators. Generative AI was becoming another member of that long list albeit a remarkably powerful one.

A year and a half later, that argument is being tested in a way I did not fully anticipate.

AI-generated content is beginning to acquire something resembling a digital fingerprint.

Anthropic has introduced machine-readable marking for supported Claude-generated content as part of its response to the European Union’s new AI transparency requirements. New supported Claude models can embed an imperceptible watermark in generated text, while supported image and file outputs can carry digitally signed provenance information using C2PA. The text watermark is intended to travel with copied-and-pasted material and may survive some editing. (Axios)

This is technically fascinating. But philosophically, it raises a much bigger question: If we can identify that AI participated in producing a piece of writing, what exactly have we proven?

Detection Is Not the Same as Judgment

Suppose you read an insightful article tomorrow. It introduces an original argument. The facts are accurate. The author connects several developments that others have overlooked. It makes you reconsider an assumption you have held for years.

Then a system tells you: AI-generated content detected.

Has the article suddenly become less insightful? I don’t believe so.

And this distinction may become increasingly important as watermarking and provenance technologies mature.

The EU’s transparency rules, applicable from 2 August 2026, are intended to make certain AI-generated or manipulated content machine-readable and identifiable. The objective is understandable: in a world of deepfakes, misinformation and industrial-scale synthetic content, knowing the provenance of digital material can be important to maintaining trust. (Digital Strategy)

Transparency is valuable. But provenance and quality are two different questions.

A watermark can potentially tell us something about how content was produced. It cannot necessarily tell us why it was produced, who conceived the argument, whether the reasoning is sound, whether the author understands the subject, or whether the content contributes anything worthwhile.

That remains a human judgment.

The Question Is No Longer “Did You Use AI?”

This is where I would update my 2025 argument. Back then, much of the debate revolved around a binary question: Did a human write this, or did AI write it?

That distinction is becoming increasingly inadequate.

Consider a professional who spends years developing expertise in a field. She identifies a new trend, develops a hypothesis, collects evidence, challenges her assumptions and determines what she wants to communicate. She then asks an AI system to help structure the argument, improve the language and turn her notes into a coherent first draft.

Compare that with someone who types: “Write me an insightful article about AI”. They copy the response without checking a single claim and publish it under their name.

Both pieces might technically contain AI-generated text. But intellectually, they are completely different acts of creation. A watermark might classify them similarly. We should not.

We Need a Better Concept of Authorship

Perhaps the arrival of AI provenance will force us to reconsider what authorship actually means.

For centuries, authorship has rarely meant physically producing every element of a finished work without assistance:

• Executives have speechwriters.
• Researchers have research assistants.
• Authors have editors.
• Journalists have fact-checkers.
• Academics use statistical software.
• Architects use CAD.
• Photographers use sophisticated computational processing.
• Programmers rely on frameworks, libraries, documentation, Stack Overflow and increasingly AI coding assistants.

The existence of assistance does not automatically transfer authorship away from the person directing the work. What matters is human agency:

• Who selected the problem?
• Who formed the argument?
• Who decided which ideas mattered?
• Who challenged the output?
• Who verified the facts?
• Who accepted or rejected the recommendations?
• And most importantly: Who takes responsibility for the final result?

These questions tell us far more about meaningful authorship than whether every sentence originated directly from a human keyboard.

AI Is Moving From Ghostwriter to Power Tool

There is another way to think about this.

Nobody asks an architect whether the building was “computer-generated” because AutoCAD was involved. Nobody dismisses a financial model because Excel calculated the formulas. Nobody accuses a photographer of abandoning photography because sophisticated software processed the image.

Why? Because society eventually learned to distinguish between the tool and the practitioner using it.

Generative AI is going through that same transition. The difference is that AI operates much closer to language, reasoning and creativity; the capabilities we traditionally associate with human intellect. That makes the boundary psychologically more uncomfortable.

• When a calculator performs multiplication, we still clearly see ourselves as the thinker.
• When an AI produces three elegant paragraphs from our instructions, the division of intellectual labour becomes less obvious.

That discomfort is legitimate. But pretending the tool does not exist or treating its use as inherently shameful—is not the solution. Developing better norms around its use is.

Watermarks Should Create Transparency, Not Stigma

This is where I believe the emerging provenance infrastructure must be handled carefully. There is a major difference between saying:

1. “AI participated in creating this content.”
2. “Therefore, this content has less value.”

The first is information. The second is a judgment. We should not confuse them.

AI provenance can be extremely useful in situations involving impersonation, deepfakes, academic integrity, misinformation, fraud or content presented deceptively as original human work. But in professional knowledge work, the situation is more nuanced.

• If an executive uses AI to transform his own strategic thinking into a clearer paper, what should the watermark mean?
• If a scientist uses AI to improve the English in a paper containing her own research, who authored the paper?
• If an engineer dictates a detailed technical argument and asks AI to restructure it, is the resulting text AI content or human expertise expressed through an AI tool?
• If I spend hours researching an idea, debating it with AI, rejecting its suggestions, adding my experience and eventually asking the system to help me articulate the conclusion—what exactly should the machine-readable label tell you about my contribution?

These are no longer hypothetical questions. They are becoming questions of digital literacy.

From Content Provenance to Intellectual Provenance

Perhaps what we ultimately need is something deeper than content provenance. We need to think about intellectual provenance.

• Content provenance asks: Where did these words or pixels come from?
• Intellectual provenance asks: Where did the thinking come from?

That distinction will become increasingly important.

• AI may generate the sentence while the human provides the insight.
• AI may generate the structure while the human provides decades of experience.
• AI may propose ten possibilities while the human recognizes the one that actually matters.
• Conversely, AI may generate both the argument and the language while a human contributes almost nothing beyond pressing publish.

Those situations should not be treated as equivalent simply because both outputs carry the same technical marker.

My 2025 Position Hasn’t Changed. But It Has Evolved

When I wrote Generative AI: A Tool, Not a Shameful Shortcut, I argued that we should judge AI-assisted content primarily through relevance, accuracy, engagement and the value it delivers.

I still believe that. But I would add another criterion today: Accountability.

Use AI. Use it extensively if it makes you better. Let it challenge your thinking. Let it organize your research. Let it criticize your assumptions. Let it help you write, code, analyze, visualize and explore.

But own the result.

• If there is an error, you cannot blame the model.
• If a source is fabricated, you remain responsible for citing it.
• If an argument is weak, saying “AI wrote it” does not strengthen it.
• And if you publish something under your name, you should be prepared to defend the ideas contained within it.

That, to me, is the boundary that matters.

The Future Is Not Human vs. AI

We are entering an Internet where digital content may increasingly carry information about how it was produced. That is probably inevitable, and in many contexts, desirable.

But we should resist allowing provenance technologies to create a simplistic hierarchy in which “human-generated” automatically means authentic and “AI-generated” automatically means inferior.

Humans can produce misinformation. Humans can plagiarize. Humans can write shallow articles. And AI-assisted humans can produce extraordinary work.

The real divide will increasingly be between people who use AI passively and people who use it intelligently:

• Between generation and judgment.
• Between automation and augmentation.
• Between those who outsource their thinking and those who use technology to extend it.

So if tomorrow one of my articles carries a machine-readable indication that AI participated in its creation, I will not consider that a mark of embarrassment. Consider it a disclosure about one of the tools in my workshop.

Then judge the article on the questions that matter:

• Is the idea worth considering?
• Are the facts accurate?
• Does the argument hold?
• Did it make you think?
• And am I willing to stand behind it?

Because the most important watermark on any piece of thought leadership should not be hidden inside its text. It should be the judgment, experience, responsibility and point of view of the human being whose name appears above it.

AI is a tool. The watermark may tell you that I used it. It still cannot tell you whether I used it well.

#GenerativeAI #ResponsibleAI #AITransparency #AIWatermarking #DigitalTrust #ContentProvenance #ArtificialIntelligence #ThoughtLeadership #AIandCreativity #HumanAI #FutureOfWork #DigitalTransformation

---

Originally published in the VerbionPulse newsletter on LinkedIn: The Watermark Changes Nothing: AI Is Still a Tool,Not the Author Revisiting Generative AI: A Tool, Not a Shameful Shortcut in the age of AI provenance.