"AI Info" Label: It's Telling on Your Creative Process

I’ve taken art classes.

More than one.

Online. In person. One-on-one.

I can tell you what value is and what hue is and how to build a composition, and I still can’t get the thing in my head onto paper. It’s my hand, and no amount of practice has moved it.

So for the last two or three years I’ve heard a particular piece of advice. Pick up a pencil. Take a class. Practice like the rest of us did. Everybody starts out bad, you just quit. You’re being lazy.

Then last month, some of the people who gave me that advice opened Instagram and found an AI info label sitting under a photo they shot themselves and edited themselves. They’re furious. Some are claiming conspiracy.

I understand the fury, and I want to walk through where it came from, because the answer is more interesting than the anger.

What changed on August 2

On 2 August 2026, the transparency rules in Article 50 of the EU AI Act began to apply.

Article 50(2) says providers of AI systems generating synthetic image, audio, video or text content have to make sure the output is “marked in a machine-readable format and detectable as artificially generated or manipulated.”

Machine-readable means the mark is meant for a computer, so most companies do it by tucking a small file inside your file. Think of the paper slip that used to come in the pocket of a new pair of jeans: inspected by number 14. It rides along, it names who touched the thing, and it’s there whether or not you read it. In an image, that slip is called Content Credentials, and it lists what made the file, what software opened it, and whether AI was involved anywhere along the way. The standard behind it is run by a group called the Coalition for Content Provenance and Authenticity, which everybody shortens to C2PA.

Meta reads the slip. Slip says AI, you get the label.

Here’s the part almost nobody’s read. Article 50(2) has an exemption written into it. The marking obligation doesn’t apply “where the AI systems perform an assistive function for standard editing or do not substantially alter the input data provided by the deployer or the semantics thereof.”

Standard editing. Your background remover was exempt the day that law passed. Your noise reduction was exempt. Your one-click sky replacement, arguably exempt. They’re writing the slip anyway, because a company facing a fine under a brand new law would rather over-mark than argue about where the line sits.

A little timeline, because it works in your favor. Meta had “Made with AI” running by May 2024, and renamed it to “AI info” on July 1, 2024, after photographers complained it was being put on ordinary retouching. Software’s been writing these slips, and Meta’s been reading them, for two years. What August 2 changed is how many programs write a slip at all.

What the AI info label can and can't tell you

The AI info label reports one thing: something in the chain was AI. It can’t tell you how much, or where in the process, or whether it mattered to the outcome.

It says the same three words about a picture a machine invented out of nothing and about a background you removed from a photo you took yourself, in your own kitchen, of your own quilt.

It’s also patchy, and photographers documented exactly how patchy two years ago. PetaPixel ran the test in June 2024. Remove a speck of dust with Photoshop’s Generative Fill and Instagram tagged the photo. Remove the same speck with the Spot Healing Brush and it didn’t. Same photo, same dust, same result on screen. The difference was that one tool wrote a full C2PA record into the file and the other one wrote nothing.

Adobe’s Generative Remove, which is also AI, didn’t trigger the label either, because it didn’t write the record.

So the AI info label is partly reporting which company was more thorough about its paperwork.

Now the part I’ve been saying for two years.

Background remover. Autocorrect. Searching your camera roll for the word “blue” and having it hand you every blue thing you’ve ever photographed. Noise reduction on a picture you took in a badly lit church basement at guild.

Most of us have been using AI every day for years. The AI info label is the first thing that ever said so out loud. I hit three of those before breakfast, and so did you.

The rule was "you could do it by hand"

Some of the people finding labels on their work this month are handling it by explaining that theirs is the acceptable kind, because theirs comes in after the picture already exists. Assistive, they’re calling it.

Here’s the trouble with that.

Every single one of those jobs could be done by hand. You could cut that background out by hand, with a scalpel and a lightbox, and people did. You could retouch by hand. You could match the color by hand. Every bit of it was done by hand for about a hundred years by people who got paid for the hours it took.

So if the rule is that you could do it yourself and therefore you should, that rule has your name on it too. It was a rule right up until it applied to them.

What’s left, once you take that rule away, is where in your creative process the machine sits. Early, late, once, forty times. That’s a preference. I’d like it called one.

The idea was always mine. The picture gets it out of my head and onto something I can work with. Then I change it, cut half of it, pick the colors and the fabrics and the size, and spend months turning it into something you can put on a bed.

What to say when someone comes at you

You’re allowed to hold your ground here. Every one of these is defensible, and none of them require you to be cruel, condescending, or anything other than open and clear about it.

“Background removal is a machine learning model. It’s been one the whole time.” Nobody hid it. “Background remover” was easier to put on a menu than “machine learning model,” so that’s what went on the menu.

“The label says AI was involved somewhere. It stops there.” It reports presence. Amount, location, and whether it changed the outcome are all outside what the slip records.

“Every one of those edits could be done by hand. People did them by hand for a hundred years.” If “you could do it yourself” is the standard, it covers background removal, retouching, and color matching too.

“Where the machine sits in your process is a preference. I’d like it called one.” Early in the process, late in the process. Both are choices about workflow.

“Apply the standard evenly or put it down.” That’s the whole ask.

Here’s the point: nobody gets to assign you a reason for using a tool.

You’re welcome to dislike generative AI. You’re welcome to keep working exactly the way you work now. Plenty of people have thought hard about this and come out somewhere different from me, and I can respect a position that holds up under its own weight.

What I want is a standard that survives contact with the person applying it.

For me, generative AI is a form of assistive AI, if that’s the name we’re giving technology somewhere in our creative process doing a thing we either can’t or won’t do by hand.

Generative AI gave people with ideas a way to get them out.

Some of us needed that.

Come learn how this works

I teach AI 101 On Demand. Twenty lessons you watch at your own pace, $29 to $49, plain language, no jargon and no doom. You’ll leave knowing what these tools do, what to worry about, what to let go, and how to spot a fake.