AI Can Label Your Work. CAHDD Lets You Tell the Human Side of the Story.

Invisible AI watermarks can tell a machine that artificial intelligence touched your work. They cannot tell another human how much of that work was actually yours. CAHDD was designed to put that information up front — visibly, voluntarily, and in human terms — as part of the first impression of your content. It gives creators a way to declare their Human Sweat Equity™ before an invisible machine-readable signal is interpreted somewhere farther down the line and reduced to a generic label that may tell only a fraction of the story.

For quite some time, we have been talking about the possibility that AI provenance, embedded identification, and invisible watermarking would eventually become part of the creative landscape.

For quite some time, many people didn’t seem particularly concerned.

Maybe the technology wouldn’t work.

Maybe people would find ways around it.

Maybe governments wouldn’t require it.

Maybe AI companies wouldn’t implement it.

Maybe you could simply strip the metadata, save something as plain text, or copy and paste it somewhere else.

And maybe none of this would ever really affect the average person using AI as another tool in their creative process.

Well, we don’t have to speculate quite as much anymore.

Anthropic has begun implementing an imperceptible watermarking system in text produced by supported Claude models. Unlike traditional metadata attached to a file, the text watermark is designed to exist within the generated language itself, making Claude’s involvement statistically detectable even after common actions such as copying and pasting or some light editing.

Anthropic is hardly doing this in a vacuum. The European Union’s AI Act transparency requirements took effect on August 2, 2026, requiring providers covered by Article 50 to make certain AI-generated or manipulated content detectable in a machine-readable manner.

And suddenly, people are paying attention.

Some Claude users are publicly questioning what happens to work they substantially created themselves but allowed Claude to proofread, edit, translate, reorganize, or otherwise process.

Questions that seemed academic when all of this was somewhere off in the future suddenly become considerably more important when it is your writing, your artwork, your design, your reputation, and your livelihood being discussed.

The conversation isn’t theoretical anymore.

And that is exactly where CAHDD enters the picture.

The Machine Can Tell Its Story. You Should Be Able to Tell Yours.

Let’s start with something important.

We are not opposed to provenance.

We are not opposed to transparency.

And CAHDD was certainly never created as a way to hide the use of artificial intelligence.

Quite the opposite.

CAHDD — Computer Aided Human Designed & Developed — was created because we believe transparency surrounding the relationship between humans and technology is going to become increasingly important.

But we also believe transparency has to work both ways.

An AI company has an understandable interest in identifying content produced or modified by its technology.

Regulators have an understandable interest in making AI-generated and manipulated content identifiable.

Platforms may eventually have an understandable interest in receiving that information and communicating it to their users.

But none of those things necessarily answer the question that matters most to the person who actually created the work:

How much of this was mine?

That is the human side of the story.

And it can be considerably more complicated than an invisible watermark can explain.

“AI Was Involved” Is Not the Same as “AI Created This”

Imagine two people writing an article.

The first spends several days researching the subject.

They develop the premise.

They organize the argument.

They write 2,000 words.

They rewrite sections that don’t work.

They change the conclusion.

They agonize over a paragraph for twenty minutes because something about it doesn’t sound right.

They finally finish the article and ask an AI assistant:

“Please correct the spelling and grammar without changing my writing style.”

The second person types:

“Write me a 2,000-word article about this subject.”

Both people used artificial intelligence.

But did they do the same thing?

Of course not.

The first article contains substantial human authorship and Human Sweat Equity™.

The second contains substantially more machine-generated authorship under human direction.

That doesn’t automatically make either approach right or wrong.

But they are clearly not the same creative process.

A machine-readable watermark may be capable of identifying AI involvement in both.

What it cannot necessarily communicate is the enormous difference between them.

And somewhere farther down the line, that distinction may disappear entirely.

A website, publisher, employer, school, social network, image-sharing service, search engine, or some platform that doesn’t even exist yet may eventually receive an AI provenance signal and translate it into something much simpler for its users:

AI Enhanced

AI Modified

AI Assisted

Or potentially:

AI Generated

That may be technically defensible from the perspective of the system applying the label.

It may also tell a completely inadequate story about the human who created the work.

That is the problem.

Their Compliance Problem Is Not Necessarily Your Authorship Problem

The companies developing AI systems are dealing with an increasingly complicated regulatory environment.

The European Union’s Article 50 requirements are a perfect example.

Providers of generative AI systems have obligations surrounding the marking and detectability of AI-generated or manipulated content. The European Commission’s guidance specifically discusses machine-readable solutions intended to be effective, interoperable, robust, and reliable where technically feasible.

Those requirements address legitimate concerns about deception, manipulation, deepfakes, and the growing difficulty of determining where digital content came from.

But satisfying that requirement doesn’t automatically solve the creator’s problem.

A machine-readable signal can help an AI provider demonstrate that its technology participated in producing something.

Great.

But what happens to that signal next?

What does another company call it?

How does an employer interpret it?

How does a publisher interpret it?

How does a client interpret it?

How does a social-media platform interpret it?

How does an automated moderation system interpret it?

How does the person looking at your work interpret it?

And what happens when a sophisticated technical distinction eventually becomes a little badge next to your work saying:

AI Modified

The AI developer may have successfully addressed its regulatory obligation.

The platform may have successfully applied its disclosure policy.

And the human creator may still be standing there saying:

Wait a minute. I actually created this.

CAHDD was designed to give that person a voice.

You May Not Be Able to Simply Strip the AI Away

One of the immediate reactions to invisible AI watermarking has been predictable.

I’ll just copy it into a text editor.

I’ll save it as ASCII.

I’ll strip the metadata.

That may fundamentally misunderstand what some forms of AI text watermarking are designed to do.

We traditionally think of digital identification as something attached to a file.

Metadata.

A hidden field.

An embedded tag.

Remove the tag and the identification disappears.

Statistical text watermarking can work differently.

Instead of hiding a conventional identifier somewhere inside the document, the model can subtly influence its choices among otherwise reasonable words or tokens as it generates text.

To you and me, the resulting language can look completely ordinary.

But across enough text, those choices can create a statistical pattern that a detector equipped to recognize it can identify.

The watermark isn’t necessarily sitting behind the words.

The words themselves can carry the pattern.

Saving those same words as ASCII doesn’t magically change them.

Copying and pasting them into another application doesn’t necessarily change them either.

That doesn’t mean these systems are infallible.

They aren’t.

Significant rewriting, paraphrasing, translation, mixing with other text, short passages, and other transformations can weaken statistical signals. Detection systems can have limitations, and both false positives and false negatives remain legitimate concerns.

But that only reinforces the larger point.

This may not become a simple game of deleting a hidden tag.

And CAHDD was never designed around the assumption that creators would somehow defeat these systems.

We designed it around something much simpler.

Tell the truth about how you created the work.

Why CAHDD Is Deliberately Low Tech

There is a certain irony in creating a deliberately simple visual system to address one of the most technologically complicated issues facing creators.

That simplicity is intentional.

We don’t know what invisible watermark Anthropic will use five years from now.

We don’t know what OpenAI will use.

We don’t know what Google, Adobe, Microsoft, Meta, or the next generation of creative platforms will use.

We don’t know which provenance standards will ultimately dominate.

We don’t know what detection systems will be developed.

And we certainly don’t know how every platform receiving that information will interpret it.

We don’t need to.

CAHDD doesn’t have to win a technological arms race with any of them.

The CAHDD ranking system is designed to do something much more human.

It allows a creator to visibly indicate the level of human authorship and technological assistance involved in creating something.

Right there.

With the work.

Humanly readable.

Humanly understandable.

And preferably visible as part of the first impression.

Before somebody wonders.

Before somebody runs a detector.

Before some platform interprets an invisible signal.

Before an algorithm places your work into a category.

You have already told them:

Here is how this was created.

First Impression Matters

We think this aspect of AI disclosure is going to become increasingly important.

Imagine seeing an illustration, photograph, article, design, architectural rendering, video, or piece of music online.

Next to it is a platform-generated label:

AI Modified

What is your immediate reaction?

For some people, that label may simply mean technology was used somewhere in the process.

For others, it may mean:

“AI made this.”

Those are very different interpretations.

But first impressions happen quickly.

Most people aren’t going to investigate which AI model was involved, which watermarking protocol was used, what confidence threshold triggered the label, whether AI changed three words or three thousand, or whether the underlying work existed before the AI ever touched it.

They are going to see the label.

That label becomes part of the work.

CAHDD approaches this from the other direction.

Instead of waiting for technology to describe your relationship with technology, you describe it yourself.

Visibly.

Voluntarily.

Up front.

The viewer sees your declaration as part of the work’s presentation.

Then, if some platform subsequently says AI was involved?

Fine.

We already told you that.

But we also told you considerably more.

This Is Why CAHDD Uses Levels

From the beginning, we rejected the idea that creative work could adequately be divided into only two buckets:

Human Created

and

AI Generated

That isn’t how modern creativity works.

Technology exists on a continuum.

A person might use digital tools while retaining essentially complete creative control.

Another might use AI for research or brainstorming.

Another might use it to correct spelling and grammar.

Another might ask it to rewrite portions of an existing document.

Another might collaboratively develop ideas with AI over dozens of iterations.

Another might generate most of the finished product through AI and primarily provide direction.

Those are different levels of technological involvement.

More importantly, they represent different levels of human involvement.

That is why CAHDD uses a Stage 0–5/X ranking system.

The purpose isn’t to shame someone for using more technology.

It isn’t to declare that a lower or higher number makes someone a better creator.

It is to provide context.

We want somebody looking at a CAHDD designation to understand that the person behind the work has voluntarily told them something about the balance between human creation and technological assistance.

Because that balance is the real story.

Human Sweat Equity™

We’ve used the term Human Sweat Equity™ because we think it describes something that conventional AI disclosure struggles to capture.

Creating something isn’t simply about who produced the final pixels or words.

It is the thinking.

The research.

The experience accumulated over decades.

The false starts.

The sketch that didn’t work.

The twenty prompts that produced garbage before one finally moved the idea forward.

The photograph you took yourself.

The model you built.

The paragraph you rewrote six times.

The judgment that told you something was wrong even when the machine insisted it was right.

The decision to throw away three hours of work and start again.

The final choice.

That is Human Sweat Equity™.

AI can participate in that process without replacing it.

And the amount of human participation can vary enormously from one piece of work to another.

A binary watermark can’t adequately measure that.

A generic platform label can’t adequately explain it.

CAHDD gives the creator a way to declare it.

We Don’t Have to Fight the System

There is another reason we deliberately kept CAHDD simple.

Most people have better things to do.

Creators want to create.

Designers want to design.

Writers want to write.

Photographers want to shoot.

Architects want to design buildings.

Musicians want to make music.

And when the workday is over, most of us would rather move on to the next project, work on the car, take the dog for a walk, or spend some time with our families than spend our evening trying to understand the latest invisible watermarking protocol.

We don’t believe creators should have to become experts in statistical watermarking, provenance standards, metadata, detection algorithms, regulatory requirements, or whatever new technical system comes along next just to defend their own authorship.

Nor do we believe we are going to stop what is coming.

Governments are going to regulate.

AI companies are going to comply.

Technology companies are going to develop increasingly sophisticated provenance systems.

Platforms are going to decide how they interpret the information those systems provide.

Some of those decisions will probably be good.

Some probably won’t be.

We can spend an enormous amount of time fighting every new system as it arrives, trying to remove invisible signatures, challenging classifications, learning another technical standard, and wading deeper and deeper into the subsurface murk.

Or we can do something remarkably simple.

Tell our story.

Put a visible CAHDD designation with the work.

Tell another human being how the work was created.

Indicate how much technology participated.

Indicate how much human authorship remained.

Claim the work we actually did.

And stand behind it.

Then move on.

There is something deliberately old-fashioned about that idea.

A human being creates something and puts their mark on it.

I made this. This is how I made it. I stand behind it.

CAHDD doesn’t require a complicated technical infrastructure to communicate that message.

There is no steep learning curve.

You don’t have to understand how the invisible watermark works.

You don’t have to know what statistical pattern an AI model may have embedded in your words.

You don’t have to understand every provenance standard operating underneath your image.

And you don’t have to defeat any of them.

Let the bureaucrats regulate.

Let the AI companies comply.

Let the machines talk to the machines.

CAHDD lets humans talk to humans.

That may sound almost ridiculously simple compared with the technology developing around us.

That’s the point.

We aren’t trying to replace machine-readable provenance.

We aren’t trying to circumvent it.

And we aren’t asking creators to fight an increasingly complicated technological system every time they finish something.

We’re simply preserving a place for the human creator to say:

Here’s what I did.

Here’s how much of this is mine.

And I’m willing to put my name behind it.

Then go create something else.

Or better yet, go spend some time with the people who actually matter.

The machines can keep talking while you’re gone.

What’s Embedded Is Inevitable. What’s Visible Is Your Choice.

When we began developing CAHDD, one of the ideas we kept returning to was that embedded provenance was probably coming whether creators wanted it or not.

That led to a phrase that has become increasingly central to what we are building:

What’s embedded is inevitable. What’s visible is your choice.

That statement feels considerably less theoretical today.

We aren’t suggesting that every AI company will use the same technology Anthropic is implementing.

They won’t.

We aren’t suggesting that every invisible watermark will survive forever.

It won’t.

And we aren’t suggesting that AI detection will suddenly become perfect.

It won’t.

We are saying something much simpler.

The direction is becoming clear.

Governments want greater AI transparency.

AI developers are building systems to provide it.

Provenance standards are developing.

Machine-readable identification is moving from theory into actual products.

And downstream platforms will eventually have to decide what to do with all that information.

Creators should have a voice in that conversation too.

This Is Only the Beginning

Anthropic isn’t the story.

Anthropic is the latest evidence of the story.

The larger story is that the distinction between human-created, computer-assisted, AI-assisted, AI-modified, and AI-generated work is becoming part of the infrastructure surrounding digital creativity.

A few years ago, it was easy to shrug that off.

It was something that might happen someday.

Someday has begun arriving.

And now people who weren’t particularly concerned about invisible provenance are suddenly asking very practical questions about what happens when their own work carries an AI signature.

We understand the concern.

But we don’t think the answer is to hide from the technology.

We don’t think the answer is to spend our time trying to outsmart it either.

We think the answer is better transparency.

Transparency that doesn’t stop with:

AI touched this.

Transparency that continues with:

Here’s what the human did.

If AI corrected your spelling and grammar, say so.

If AI helped you research, say so.

If AI helped develop an idea you then created yourself, say so.

If you worked collaboratively with AI throughout the process, say so.

If AI generated most of the finished work under your direction, say so.

And if the work represents enormous amounts of human thought, skill, experience, judgment, experimentation, and plain old-fashioned work, you should be able to say that too.

We don’t know what invisible marks tomorrow’s technology will place inside our creations.

We don’t know what generic labels future platforms will attach to them.

And, like the enormous mass of an iceberg hidden beneath the surface, we probably don’t yet know the full scale of what is developing underneath the digital tools we use every day.

We don’t have to.

We can acknowledge that it is there without spending our lives swimming around underneath the surface trying to map every inch of it.

Because we do know one thing.

A machine-readable watermark may be able to tell the world that artificial intelligence touched your work.

CAHDD was designed so you can tell the world how much of that work was still yours.

Claim your authorship.

Declare your Human Sweat Equity™.

Stand behind your work.

Then get on with being human.

Computer Aided Human Designed & Developed™

Creator choice, before platforms decide for you.

What’s embedded is inevitable. What’s visible is your choice.