Qualification: The Right to Contest AI
An explanation is not a challenge procedure. Kaminski and Urban's account of contesting AI separates information that makes a decision intelligible from the process that permits an affected person to challenge it. Their framework starts with notice and reasons, then adds access to evidence, an opportunity to be heard, and review by a legitimate decision-maker. The institutional test is therefore not whether a system records or explains a denial. The test is whether a person can invoke a defined process that reaches an actor with authority to reconsider the decision and provide a new outcome. This distinction qualifies the Mosaic finding. The paper does not inspect Mosaic Companion, Gatekeeper, or commit 2d920ce. It does not establish that every agent decision requires judicial process or a neutral external tribunal. It also leaves open how much explanation or review a particular risk level warrants. Its contribution is narrower. A reason string and an audit entry may supply notice and grounds. They do not by themselves create standing, a hearing, a decision rule, or remedial authority. A sovereign local agent runtime should bind those elements before a consequential decision is enforced. The challenge record should identify the contested decision, the grounds and evidence available to the affected party, the reviewer, the review standard, and the set of authorized remedies. If the reviewer cannot suspend, revise, or reverse the relevant allocation or authority state, the system has an explanation channel, not a contestation mechanism.
governance-designinstitutional-designai-and-agentsaccountabilitycontestabilitydue-processhuman-reviewremedy