AI Gatekeepers

The paper gives an overview on what established gatekeeper theory on financial markets was designed to achieve. It moves on to introduce three groups of AI-gatekeepers: established auditors, new entrants, and publicly accredited gatekeepers. Then, the paper describes designated AI-laws and designated AI-gatekeepers, both optional and mandatory. It concludes with a preliminary roadmap for regulating AI-gatekeepers.

The paper makes three contributions to the debate. First, it shows why a transplant of the proposal to introduce a version of strict gatekeeper liability from the financial-market setting is inadequate for AI-gatekeepers. Second, it suggests a two-layer transparency regime: aggregate disclosure of testing methodologies and standards, combined with compact certifications or scores at the individual level. Third, the paper proposes ex ante governance requirements operating at two sites: At the gatekeeper itself, through accreditation and independence requirements modelled on the EU's notified-body regime. Additionally, at the interface of the gatekeeper with the client, through institutional structures analogous to the audit committee but designed for the interdisciplinary translation problem that distinguishes AI-compliance review from classical audit.