A good deal of the commentary on the transparency obligations has been written for European readers, which has left a gap. Article 50 applies to any AI system placed on the EU market, and it also applies to providers and deployers located outside the Union where the system’s output is used inside it. A firm with no European entity, no European office, and no European infrastructure can be in scope on the strength of where its work ends up.
For professional services this is less exotic than it sounds. A New York consultancy writing a market analysis for a client with a Dutch subsidiary, a London agency producing campaign copy that runs in Germany, a research firm whose memo gets circulated to an investment committee in Frankfurt: in each case the output is used in the Union while the firm producing it sits outside. Geography of the business and geography of the output are different things, and the regulation follows the second one.
The usual mistake is to reason from corporate structure. No European entity, therefore no European obligations, therefore this is a problem for next year or for somebody else. That logic works for a great deal of regulation and does not work here, which is exactly why it catches people: the reasoning is habitual rather than careless.
The second mistake is subtler. Firms assess exposure by looking at their direct clients and stop there, when output travels further than the engagement does. A deliverable handed to a US client gets forwarded to that client’s European counsel, attached to a board pack, or quoted in a document that circulates across the group. Where a professional firm’s work ends up is frequently not something the firm controls or even observes.
It does not mean every firm with an occasional European reader is in scope of everything. Whether a particular arrangement triggers a particular duty depends on specifics: what the system does, how output is used, who the parties are, and which of Article 50’s four duties is in question. That is a legal determination that depends on facts a blog post cannot see, and it is the reason this is a conversation for counsel rather than a conclusion you can reach from an article.
What it does mean is that “we are not a European company” has stopped being a complete answer, and firms using it as one should at least know they are relying on something weaker than they think. The useful posture is to find out where your work actually lands before somebody else tells you.
There is a version of this problem that does not require resolving the jurisdictional question at all, which is why it is worth leading with. If you can show, for any piece of AI-assisted work, what evidence it rested on and who reviewed it before it went out, the question of which regime applies becomes a matter of describing a process you already run rather than constructing one under time pressure.
That capability is also not jurisdiction-specific in any direction. Clients in New York, London and Frankfurt are converging on the same question about AI-assisted work, with different vocabulary and different legal hooks behind it. A firm that built the answer once has answered all of them, which is a considerably better position than maintaining three separate stories about the same workflow.
Qonera is the AI governance platform for professional teams, built around a structured review and approval workflow: evidence checked before analysis, claims cited back to source, independent models stress testing each other, and a named reviewer approving before delivery, with the whole sequence captured in a tamper evident audit trail. Our hosting, data residency and security commitments are published on the trust page. Where your firm is incorporated turns out to matter less than where your work is read, and most firms have never checked the second one.
This article is for general information only and does not provide legal advice. Organisations should consult qualified legal counsel about how Article 50 and the EU AI Act apply to their specific systems, workflows, and obligations, including questions of territorial scope.
Multi-model stress testing, Conflict Heatmap, tamper-evident audit trail, and structured sign-off, built for teams who need defensible AI output.