Library
White paperMichael Clark
Literacy as a Control, Not a Course
AI oversight depends on the overseer's competence. Many programmes anchor on Article 4, which from 27 July 2026 asks only for measures to support AI literacy, but Article 26(2) is the provision that will test competence, from 2 December 2027 for Annex III systems, subject to Article 111(2) for systems already on the market. SM&CR, NIS2, DORA and the AI Act ask for competence records with the same shape, and the recommended control is a dated, per-person competency record tied to decision types, with sign-off authority that lapses when the record does.
Key findings
- Every oversight duty assumes the overseer can evaluate what they are shown, and without that competence oversight fails as a rubber stamp, an override or a nominal review.
- Article 4 asks for effort; Article 26(2) asks for competence necessary for the particular system, and it is testable person by person.
- A completion rate cannot show that an overseer is competent for the system and the decision.
In the paper
- Three ways oversight fails
- Article 4 asks for effort
- Article 26(2) asks for competence
- A completion rate cannot show competence
- Competency should gate authority
- One register, four regimes
- Where to start
Michael Clark
Chief Executive Officer, Standard Intelligence
Your next step