AI systems are increasingly used to support institutional decisions affecting education, welfare, immigration, healthcare, employment, finance, professional regulation and public administration. Existing accountability mechanisms assume that a decision can be challenged if reasons are given, a human is involved, and an appeal or complaint route exists. That assumption is no longer safe. Where an AI system materially influences a decision, the real basis of the decision may be embedded in a model, confidence score, threshold, data pipeline, vendor-controlled process or automated workflow. The affected person may receive reasons and access to review, yet still be unable to test the criteria, evidence or technical basis of the decision. The institution may not fully understand the system. The reviewer may lack technical competence. The vendor may invoke confidentiality. The regulator may have only partial jurisdiction. This brief calls that failure the capture of answerability: the condition in which a decision is formally accountable but practically unchallengeable. The policy response should not be limited to transparency or human-in-the-loop safeguards. Legislators and regulators should require a minimum answerability standard: functional disclosure, competent review, procurement-stage auditability, traceable responsibility, effective remedies, and a Law Commission review of liability allocation in AI-mediated public and quasi-public decision-making. This brief draws on the underlying working paper: Peter Kahl, ‘The Capture of Answerability: AI Governance and Institutional Accountability’ LXR-AIGOV-ANS-2026-01, Lex et Ratio Ltd Working Paper, 9 May 2026, DOI: 10.5281/zenodo.20093912.
Peter Kahl (Sat,) studied this question.