Right to Send Your Agent: Citizen Power in the AI Services Economy

The strongest objection is that this could become fraud infrastructure. If agents can act at machine speed across banks, benefits, health care, utilities, and courts, a compromised agent could drain accounts, expose medical records, file false claims, harass institutions, or trap vulnerable people inside coerced consent. A second objection is that public agencies and small providers already struggle to maintain ordinary systems; forcing them to accept agent traffic could add cost, attack surface, and service risk.

Those objections are strong enough to shape the policy. The right should not be implemented as permissionless bot access. It should be implemented as verified delegation: agent identity, least-privilege scopes, revocation, receipts, audit trails, anomaly detection, high-risk confirmation, incident reporting, account recovery, and human fallback. It should be phased where public rails already exist and expanded only when conformance tests and fraud controls work. The position holds because the alternative is not a low-risk human world. It is a world where institutions automate first and citizens are left with passwords, PDFs, and exhaustion.

Federal agencies, regulated firms, standards bodies, and technology vendors bear the cost of APIs, identity integration, conformance testing, audit logs, fraud monitoring, and incident response. Agent providers bear the cost of security engineering, user support, model evaluation, explainability, and liability coverage. Legal aid and public-interest organizations bear implementation burdens if the public option is underfunded. Small providers may need phased deadlines, safe harbors, shared tooling, or exemptions until common infrastructure reduces cost.

Citizens also bear risk if the design is wrong: coerced authorization, confusing consent, digital exclusion, mistaken filings, lost privacy, or overreliance on automated advice. Those costs are acceptable only if the policy includes plain-language consent, revocation, paper alternatives, human escalation, disability and language access, and limits on high-stakes autonomous action.