{"resourceId":"aws-agentcore-human-in-the-loop-snap-medicaid-eligibility-20260901","versions":[{"version":"external-d82e7b173a4e8428c81a1f8aec3252460927ceb6894f3c77784546c9b953cbae","resource":{"id":"aws-agentcore-human-in-the-loop-snap-medicaid-eligibility-20260901","title":"AWS outlines a human-in-the-loop autonomy model for agentic SNAP and Medicaid eligibility processing on Bedrock AgentCore","organization":"Amazon Web Services (AWS Public Sector Blog)","sector":"Health and human services (state eligibility programs)","geography":"United States","publishedAt":"September 1, 2026","publicationDate":"2026-09-01","eventDate":null,"sourceName":"AWS Public Sector Blog","sourceLabel":"Vendor architecture blog post","sourceUrl":"https://aws.amazon.com/blogs/publicsector/human-in-the-loop-claims-processing-with-amazon-bedrock-agentcore/","evidenceClass":"vendor-claim","outcomeClass":"emerging","topics":["developers-agents","governance-procurement","operating-model","data-security"],"finding":"AWS argues most production eligibility agents should not start fully autonomous, noting states face sanctions when SNAP determination error rates exceed 6%. It proposes three modes: Assist (agent drafts, caseworker approves each decision), Supervised autonomy (agent handles routine cases; edge cases, denials and low-confidence cases go to humans) and selective Full autonomy for case types with sustained accuracy. Cedar policies declaratively define what the agent may do, per program, such as authorizing SNAP determinations while keeping Medicaid recommend-only. Knowledge Bases, AgentCore Memory, Runtime, Gateway and Guardrails provide policy retrieval, case context, isolation, tool access and PII protection.","sledRelevance":"Directly applicable to state and county HHS agencies weighing agentic AI in eligibility, where error rates carry federal financial consequences.","evidence":"Vendor architecture guidance; no named state deployment or measured accuracy results. The SNAP 6% threshold is cited by AWS.","architectureImplications":"Separates authorization (Cedar policies) from agent logic; routes tools through AgentCore Gateway to case management, wage databases and federal verification hubs; uses isolated runtime and session memory.","governanceImplications":"Autonomy expands by evidence of measured accuracy, with versioned, auditable policies; denials and edge cases stay with humans.","securityPrivacyImplications":"Cedar limits case access to caseloads; Guardrails aim to prevent PII leakage and prompt injection; access to federal data hubs carries its own compliance obligations.","caveats":"Vendor guidance, published about two months before this edition; no field results; federal program rules and state policy determine what any autonomy level can legally do.","streamIds":["aws","health-human-services"],"roles":{"sales":"Sales framing: HHS leaders are wary of autonomous eligibility agents, and this guidance meets them where they are: start with the agent drafting and caseworkers approving every decision, then expand only where measured accuracy supports it. Use the SNAP error-rate sanction context to frame risk honestly. Position engagements around an Assist-mode pilot with accuracy baselining, and avoid implying any state has achieved autonomous determinations, because the post names no deployment and reports no results.","engineering":"Engineering framing: express agent authority as Cedar policies separate from agent code, scoped per program and caseload, so rules can be versioned and audited. Use Knowledge Bases for federal regulations, state manuals and county procedures, AgentCore Memory for case context, an isolated Runtime for reasoning, and Gateway for calls to case management and verification systems. Route every denial and low-confidence case to a human, and capture decision evidence so accuracy can be measured by case type.","delivery":"Delivery framing: begin in Assist mode with caseworker approval on every draft, and collect accuracy against quality-control review to establish a baseline. Define in advance the evidence needed to move a case type to supervised autonomy and who signs off, including program policy staff and counsel. Train caseworkers to review agent drafts critically. Report error rates by program regularly, and keep a documented process to pull authority back if accuracy degrades."},"roleDetails":{"sales":{"customerProblem":"Eligibility caseworkers face backlogs while error rates carry federal financial consequences.","stakeholders":["State HHS secretary or deputy","SNAP and Medicaid program directors","Eligibility operations managers","Quality control staff","Agency counsel and CISO"],"discoveryQuestions":["What is your current payment and determination error rate by program?","Which case types are most routine?","Who must approve changes to agent authority?","Which systems would the agent need to call?"],"valueHypothesis":"An Assist-mode agent may shorten caseworker drafting time while preserving accuracy, to be validated against quality-control review.","potentialEngagement":"An Assist-mode pilot on a narrow case type with accuracy baselining and policy-as-code design.","unsupportedClaims":["Accuracy or error-rate improvements","Any state operating autonomous determinations","Federal approval of autonomy levels"]},"engineering":{"fit":"Fits agentic drafting with human approval on Bedrock AgentCore.","architectureIntegration":"AgentCore Runtime, Memory and Gateway; Knowledge Bases for policy; Cedar for authorization; integration with case management and verification hubs.","prerequisites":["Policy manuals in retrievable form","Case management API access","Quality-control baseline data"],"constraints":["Federal program rules","Data-sharing agreements for verification hubs","Denials must remain human"],"security":"Cedar caseload scoping, Guardrails for PII and prompt injection, full decision audit logs.","proofOfValue":["Draft accuracy versus caseworker decisions","Zero policy-violating agent actions in testing","Time saved per case in Assist mode"]},"delivery":{"work":"Deploy an Assist-mode agent on one routine case type and measure accuracy.","dependencies":["Program policy staff","Eligibility system vendor","Security and privacy office"],"ownership":"State HHS eligibility operations with IT.","skillsAdoption":"Train caseworkers to review agent drafts and log disagreements.","governanceCheckpoints":["Policy-as-code review by program and legal staff","Accuracy gate before any autonomy expansion","Periodic error-rate reporting"],"acceptanceCriteria":["Proposed: agent drafts match final determinations at an agreed rate","Proposed: all denials human-decided","Proposed: auditable decision trail for every case"],"risks":["Erroneous determinations harming families","Automation bias among reviewers","Federal compliance exposure"]}},"retrievedAt":"2026-10-29T22:03:10-05:00","enrichedAt":"2026-10-29T22:16:30-05:00","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Caseworkers shift from drafting to reviewing and handling exceptions; training on reviewing agent drafts is needed.","procurementImplications":"Statements of work should tie autonomy levels to measured accuracy gates and require policy-as-code artifacts the state owns.","operatingModelImplications":"Requires ongoing measurement of error rates by case type and a formal process for expanding or retracting agent authority.","sourceVerification":{"openedUrl":"https://aws.amazon.com/blogs/publicsector/human-in-the-loop-claims-processing-with-amazon-bedrock-agentcore/","referenceExcerpt":"This can apply to decisions such as authorizing the agent for making SNAP determinations, but making it recommend-only for Medicaid.","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}