WHO GETS PAID WHEN AI USES YOUR WORK? AXM IS BUILDING THE PLUMBING

Every answer an AI hands you was built on somebody's work. A newspaper archive, a manuscript, a publisher's back catalog, a photo collection someone spent decades keeping. Almost none of those people know it happened, none of them are named in the output, and none of them see a cent.

A company called AXM wants to change the middle part first. Get the naming right, it argues, and the paying part follows.

Attribution Is Becoming Infrastructure

The fight over AI and creative work has mostly played out in court. Newspapers sue, artists sue, publishers sue, and every case argues the same thing after the fact: you used our stuff without asking.

AXM is starting from a different place. Instead of fighting over data after a model has already swallowed it, the company builds a layer that sits between the archive and the AI, and checks the terms on the way in. Who owns this. What are they allowing. Does the output that comes out the other end still carry their name.

That is the whole bet. Attribution stops being something you prove in a lawsuit and becomes something the pipeline does automatically, like a permission check or a payment rail. The company frames its own work as digital public infrastructure for authorship, plumbing rather than a product.

What AXM Actually Does

AXM does not train its own models, which is the part that makes it unusual. It structures how datasets get fed into other people's AI systems.

An institution loads its collection through AXM with the rights attached. From there the system verifies usage conditions in real time, keeps attribution stitched to the material as it moves through training and into generated answers, and logs the whole chain so it can be audited later. It plugs into the retrieval setups most AI products now use to pull in outside sources, and it can filter a dataset down to only what the owner actually cleared.

Payment is the optional last piece. When value gets generated off a collection, AXM can route a share back to whoever contributed it. The company calls this automated value routing. In plainer terms, the archive gets a cut, and there is a record of why.

The material it is built for is exactly the kind that usually gets scraped and forgotten: historical newspapers, manuscripts, cultural artifacts, publishing catalogs, and the rights and provenance data that says who owns what.

Why The African Union And Two HBCUs Are Building This

The names behind AXM are the tell. This is not a Silicon Valley licensing startup.

The project is being built with the African Union Development Agency, Howard University Law School's Artificial Intelligence Initiative, and the Cultural IP Research Hub at Miles College, and it has been recognized by the United Nations Industrial Development Organization. It sits in the ITU's AI for Good playbook as a live pilot, with early deployments in a university dataset licensing sandbox, a publisher catalog pilot, and a historical newspaper archive.

That coalition points the story somewhere most AI copyright coverage does not go. The people most likely to have their work used without credit are also the people least equipped to sue over it: small archives, cultural institutions, and knowledge from the Global South that gets treated as free raw material. One of the group's own papers puts it directly in the title, calling cultural sovereignty a form of infrastructure.

The framing there is that a nation's newspapers and archives are national assets, and letting them flow into foreign AI systems uncredited and unpaid is an economic problem, not just an ethical one. AXM is pitching itself as the tooling that lets an institution say yes to AI without giving itself away.

Credit As A Standard, Not A Lawsuit

The obvious catch is adoption. A permission layer only works if the AI systems on the other side agree to read it, and nobody building frontier models is under any real pressure to route their training data through a consent check today.

But the direction is worth watching, because it reframes the whole argument. The current AI-and-creators fight assumes attribution is a favor, something a platform might grant if shamed into it. AXM treats it as infrastructure, something you build once and everything runs through, the same way payments and identity got standardized.

If that version wins, the question stops being whether AI used your work. It becomes whether it read the terms first. For every institution sitting on a collection it has never been able to protect, that is the difference between being a resource and being a rights holder.