The European Commission published the three draft chapters of its Article 6 guidelines on Tuesday May 19, opening a stakeholder feedback window that runs through Monday June 23. [1] Article 6 is the operative provision of the AI Act under which a system is or is not "high-risk" — the classification that triggers conformity assessment, registration in the EU database, human-oversight obligations, and the post-market monitoring regime that will, once enforced, structure compliance budgets across Europe. The guidelines that arrived Tuesday determine what falls into each of the eight Annex III categories. The paper's Saturday coverage of the political timetable treated the publication as narrowing.

The Swiss compliance consultancy Modulos AG published its first substantive technical analysis Wednesday May 20 and reached a different conclusion. [2] Modulos's reading is that the Commission's text rotates the high-risk perimeter rather than contracting it. Some classifications previously expected to be in scope are now excluded — notably certain general-purpose biometric authentication systems where the operator does not retain the biometric data — but several categories previously interpreted narrowly are now drawn broadly, with the practical effect that more large enterprises fall inside the regime than outside it. The Saturday standard on the narrowing-versus-broadening split flagged the employment-scope expansion; the Wednesday Modulos paper extends the analysis to two further categories.

Point 4, employment-related systems, is the largest single rotation. The Commission's guidance now includes within scope any AI system used in the recruitment, screening, or performance evaluation of workers — including the candidate-ranking and resume-parsing systems that the previous narrower interpretation had treated as decision-support tools outside the regime. [2] An enterprise running a third-party applicant tracking system with embedded scoring features is, under the new reading, the deployer of a high-risk AI system and subject to the deployer obligations, including the right of an affected worker to seek meaningful information about the decision. The published guideline does not allow contracting out of the deployer-status; the obligation attaches to the use rather than to the build.

Point 5(c), the insurance-related provision, is the second rotation. The guideline draws insurance underwriting and pricing for life and health policies into scope, including the algorithmic-pricing tools the U.S. and European insurers have deployed at scale over the last five years. [2] The Commission has clarified that even systems used solely to flag applications for human underwriter review fall inside the regime when those flags meaningfully shape the underwriter's outcome. This is the technical question European insurance lobbies will spend the feedback window contesting. Their objection will be that the human-in-the-loop carve-out is being read out of the statute.

Point 3, education and vocational training, completes the trio. The guidance now includes the AI systems used to assign students to programs, score examinations, or determine access to educational opportunities — categories that capture the major learning-management systems sold to European universities and the credentialing platforms used by employer-side training providers. [2]

The political timetable agreed at the EU's May 7 trilogue extension remains operative: Annex III obligations now apply from December 2, 2027; Annex I from August 2, 2028. [3] The August 2, 2026 statutory deadline written into the original AI Act remains legally in force pending enactment of the Digital Omnibus package that would formally extend it. European compliance officers therefore face the operationally awkward position of preparing for an August 2026 deadline they expect to be moved, against guidelines whose final shape will not be settled until the June 23 feedback close. The rotation Modulos identified means the population of enterprises that should now be drafting deployer policies is larger by Sunday than it was Tuesday morning. [4]

-- HENDRIK VAN DER BERG, Brussels