The AI–Law Boundary Problem: Why Systems That Reason Like Lawyers Cannot Be Legal Actors"

This paper presents the first complete doctrinal architecture for determining the legal status of artificial intelligence systems in U.S. and EU legal contexts. Drawing exclusively on binding statutes, case law, professional rules, the AI Act, the GDPR, and the Lawyers’ Directives, the paper demonstrates that the central difficulty posed by AI is not technological but classificatory: modern AI systems exhibit formal characteristics of legal reasoning without possessing the legal subjectivity that doctrine requires for the exercise of legal judgment. The analysis shows that: AI systems can generate outputs indistinguishable from legal analysis, yet they cannot satisfy the structural conditions of agency, accountability, or professional authority, and misclassification arises when courts infer legal agency from the form of AI output rather than from the legally relevant attributes of the actor. The paper unifies four doctrinal axes: U.S. functional UPL doctrine (self‑representation vs. legal practice) Agency theory and the execution/judgment distinction Epistemic trust vs. institutional legitimacy EU’s three‑layer regulatory architecture (AI Act, GDPR, Lawyers’ Directives) The result is a boundary theorem: Legal classification turns not on intelligence, accuracy, or reasoning quality, but on who exercises legally cognizable judgment, for whom, and with what directive effect. This work does not claim to offer a final answer or propose new legal categories. It is an academic framework — a map of the existing doctrinal terrain — intended for courts, regulators, scholars, and practitioners confronting the rapidly emerging class of AI systems whose outputs resemble legal reasoning but lack the legal subjectivity that doctrine presupposes.