Public Value Principles

The technical standards in Parts One through Five serve a set of public value principles. These are design constraints that govern how the technical standards are interpreted and applied.

6.1 Equity of Access

Agent-mediated public services must preserve and extend equitable access across all populations. The introduction of AI agents into service delivery must be evaluated against its effect on access equity.

• Agent interfaces must function across varying levels of technical sophistication, device types, and connectivity.
• Services must remain available through non-agent channels. Agent-mediated access is an addition to existing channels.
• Outcomes must be monitored for disparate impact across demographic groups, geographies, and levels of digital access.
• Where agent-mediated access produces better outcomes (faster processing, fewer errors), those benefits must be extended equitably across all populations.

6.2 Clarity for the Public

People interacting with public services have the right to understand what is happening with their case, who or what is making decisions, and what their options are at every stage.

• When an agent is involved in processing a case, the person must be informed — clearly and in plain language — that an automated system is acting.
• Determinations must be explainable in concrete terms: what data was used, which rules applied, and what the outcome means.
• Available actions — including appeals, requests for human review, and complaint mechanisms — must be presented proactively.
• Status information must be current, accurate, and accessible through straightforward channels.

6.3 Protection of Human Judgment

Agents augment human judgment. The standard explicitly preserves human decision-making authority at defined points in every process.

• Adverse determinations — denials, reductions, terminations — require human review before they take effect.
• Caseworkers retain the authority to override agent determinations based on professional judgment, with the override and its reasoning recorded in the audit trail.
• Agent recommendations must be presented as recommendations. The interface must make clear that a human is making the final determination.
• Systems must monitor for automation bias — the tendency of reviewers to defer to agent recommendations without independent evaluation — and implement countermeasures.

6.4 Continuous Review

Standards evolve alongside the systems they govern, the populations they serve, and the technology they regulate. Continuous review is a structural requirement.

• Outcome data must be collected and analyzed on an ongoing basis to identify patterns of error, bias, or unintended consequences.
• Affected communities must have a defined channel to report problems, request changes, and participate in standard revision.
• Standards must be versioned and updated through a transparent process that includes public comment periods.
• Each revision must document what changed, why it changed, and what evidence supported the change.

Technology changes. Populations change. Policy changes. The commitment to public value endures — and the standards that protect it must evolve continuously.