The EU just turned disclosure into an engineering job
The European Union is applying to AI-generated text the same labelling standards it set for AI-generated video. Read quickly, that sounds like a formatting rule, the digital equivalent of a nutrition label. Read slowly, it is a provenance rule. You cannot label what you cannot trace, and tracing is a property of systems, not of policies.
Here is what most people skimming the headline will miss. Video is a small number of large artefacts, and a company can label those by hand if it must. Text is the opposite. Text is the product description, the support reply, the job advert, the internal summary, the client report, the thousand small outputs a modern company emits every day, most of them now passing through a model somewhere along the way. The obligation does not land on a content team. It lands on every system in the building that produces a sentence.
Disclosure is not a policy document. It is a property of your systems, and someone in the building has to own it.
That is a repricing, and it is a precise one. The person this market now wants is not the policy author who drafts the AI governance framework, and not the prompt specialist who produces the output. It is whoever can stand behind an arbitrary paragraph a customer has already read and say which model, which prompt, which data and which human edit produced it. That is data lineage work with legal literacy attached, and it is scarce because the two halves have rarely lived in the same head. The title barely exists yet in most organisations, which is why the job adverts for it will be written badly for another year.
I also expect the requirement to travel further than its legal scope. Clients tend to ask their vendors for the same assurance they owe their own regulator, so a rule written for Europe becomes a procurement question in places the rule never reaches. Procurement questions arrive long before enforcement does, and they arrive addressed to whoever answers the phone.
For hiring leaders, the practical read is simple. Before you buy a compliance layer, ask whoever owns your content pipelines one question: for any paragraph a customer saw last week, can you show what produced it. If the answer is no, your next platform hire should be someone who has shipped lineage and audit trails, not someone who has only ever shipped throughput.
Planning a senior AI or data hire and want the market view first? We share it either way.
Send us the brief →