{"resourceId":"connecticut-comptroller-ai-health-plan-protections-20260916","versions":[{"version":"external-aff92835b26a9929de782720883e8fb56741d992bbe6959203fc7b452ff23700","resource":{"id":"connecticut-comptroller-ai-health-plan-protections-20260916","title":"Connecticut Comptroller sets AI human-review and no-train rules for state health plans","organization":"Connecticut Office of the State Comptroller","sector":"State government / employee and public-sector health benefits","geography":"Connecticut, United States","publishedAt":"September 16, 2026","publicationDate":"2026-09-16","eventDate":"2026-09-16","sourceName":"Connecticut Office of the State Comptroller","sourceLabel":"Comptroller press release PDF","sourceUrl":"https://osc.ct.gov/wp-content/uploads/2026/09/2026.09.16-Comptroller-AI-Healthcare-Policy.pdf","evidenceClass":"government-evaluation","outcomeClass":"emerging","topics":["governance-procurement","data-security","accessibility-workforce","operating-model"],"finding":"On September 16, 2026, Comptroller Sean Scanlon announced five AI policy pillars for more than 270,000 members of the State Employee Health Plan and Partnership Plan, effective January 1, 2027: no adverse determination solely by AI without human review; no sole AI basis to downcode claims, reduce payments, or alter billing codes without human review (authorized contractual automation remains permitted); member data may not train other AI models; disclosure when AI materially assists or interacts with members; and carrier validation plus disclosure of governance/audit procedures for accuracy, consistency, and fairness. Scanlon intends to seek 2027 legislation extending the rules to all state-regulated plans. Corroborating CT Mirror reporting notes Anthem, Cigna, Aetna, and Caremark agreement and that only fully insured plans fall under state regulation (~220,000 people / 6% of residents).","sledRelevance":"Direct state-government benefits-administration AI controls for workforce and municipal partnership plan members. Transferable as a procurement/contract attachment pattern for other states, with limits: Connecticut plan authority, effective date still future, and self-funded ERISA plans remain federally regulated.","evidence":"Official OSC press release PDF dated September 16, 2026 listing five pillars, membership scope, Jan 1 2027 effective timing, and stated 2027 legislative expansion intent. CT Mirror same-day reporting corroborates carrier names, negotiation timing, and jurisdictional limits of state-regulated plans.","architectureImplications":"Requires claim and prior-auth workflows that keep humans in the loop for adverse actions; separates permitted contractual automation from prohibited sole-AI downcoding; implies disclosure UX when AI assists member interactions; requires carriers to expose validation and audit artifacts to the Comptroller.","governanceImplications":"Comptroller-owned policy for plans he administers creates an enforceable contractual gate before broader statute. Expanding to all state-regulated plans needs 2027 legislation; self-funded plans remain outside state reach.","securityPrivacyImplications":"Explicit ban on using plan member data to train, develop, or support other AI models; requires secure protection of member data. Does not specify technical isolation (vault vs contractual promise) or audit sampling methods.","caveats":"Policy announcement, not a measured outcome study. Effective date is future (2027-01-01). Enforcement depth, audit methods, and fairness metrics are not defined in the release. Coverage does not extend to most privately self-funded plans.","streamIds":["state-government"],"roles":{"sales":"Sales framing: Connecticut shows a benefits-office path to bind carriers on human review, disclosure, and no-train rules without waiting for a full insurance code rewrite. Engage state comptrollers, benefits administrators, and health-plan procurement leads who need AI denial/downcoding controls for employee and partnership plans. Position a bounded policy/contract attachment review—not guaranteed claims savings. Unsupported: that Jan 2027 rules already reduce wrongful denials statewide.","engineering":"Engineering framing: Fit is workflow and vendor-control design for claims, prior authorization, and member-facing assistants under a mandatory human-in-the-loop rule for adverse actions. Integrate decision audit trails, disclosure flags when AI assists members, and contractual plus technical blocks on using plan data for external model training, while preserving authorized automated claims processes the policy still permits. Prerequisites include adverse-action taxonomies, fairness validation datasets, and reviewer staffing models. Constraints: plan-specific Comptroller authority; technical isolation of training data is stated as policy rather than a published architecture.","delivery":"Delivery framing: Work is carrier contract amendments, disclosure UX, validation reporting to the Comptroller, and readiness rehearsals before the January 1, 2027 effective date. Ownership sits with the benefits office plus named carrier compliance leads for medical, dental, Medicare Advantage, and pharmacy. Acceptance should require demonstrated human review on adverse samples, attested no-train controls with subprocessor lists, and live member disclosures—not press-release concurrence alone. Risks include checkbox fairness reporting and uneven reviewer quality across carriers."},"roleDetails":{"sales":{"customerProblem":"State employee and partnership health plans face rising AI use in claims and prior authorization without clear human-review, disclosure, or member-data training bans.","stakeholders":["State Comptroller / benefits administrator","Employee and municipal plan sponsors","Medical/dental/pharmacy carriers","State medical society and patient advocates"],"discoveryQuestions":["Which adverse actions today can be taken without a documented human reviewer?","Where does member PHI flow into carrier or vendor models for training?","What disclosure language appears when AI assists member interactions?","Which fairness/accuracy validations can carriers produce before Jan 1 2027?"],"valueHypothesis":"Binding carriers to human review, disclosure, and no-train rules may reduce unchecked automated denials/downcoding risk for covered members if audits verify compliance—without proving better clinical outcomes.","potentialEngagement":"A 6–8 week benefits AI control assessment: map adverse-action paths, draft contract riders mirroring CT pillars, and design Comptroller-facing validation reporting.","unsupportedClaims":["Guaranteed reduction in wrongful denials","Statewide protection for all Connecticut residents before 2027 legislation","That carrier agreement equals independently verified fairness"]},"engineering":{"fit":"Strong fit for claims/prior-auth decisioning and member chat under human-review and disclosure constraints; weak fit where the buyer lacks plan-administration authority.","architectureIntegration":"Insert human-review queues for adverse determinations; tag AI-assisted member communications; enforce contractual and technical blocks on using plan data for external model training; retain authorized automated claims processing where contracts allow.","prerequisites":["Adverse-action taxonomy and reviewer staffing model","Carrier data-flow diagram including subprocessors","Validation datasets for accuracy/consistency/fairness","Disclosure copy and logging standards"],"constraints":["Effective date still future","Self-funded plans outside state regulation","Release lacks technical audit protocol detail"],"security":"Treat member PHI as prohibited training input; require encryption, access control, and retention limits; prefer verifiable isolation over promise-only clauses where feasible.","proofOfValue":["Sample audits showing human sign-off on adverse AI-assisted decisions","Attestations and technical tests that member data is excluded from training corpora","Published fairness metrics across populations as advocates requested"]},"delivery":{"work":"Negotiate carrier amendments, implement disclosure and review workflows, stand up Comptroller reporting, and run pre-2027 compliance dry-runs.","dependencies":["Carrier system changes and legal sign-off","Comptroller audit intake process","Communications to 270,000+ members"],"ownership":"Office of the State Comptroller benefits leadership with carrier compliance owners.","skillsAdoption":"Train claims reviewers and member-service staff on when AI assistance requires disclosure and escalation.","governanceCheckpoints":["Contract rider execution","Pre-effective-date validation report acceptance","Quarterly fairness/accuracy disclosures"],"acceptanceCriteria":["Documented human review on 100% of sampled adverse determinations","No-train clause attested with subprocessor list","Member-facing AI disclosure live in production channels"],"risks":["Carriers meet letter of policy without meaningful review quality","Legislative expansion stalls, leaving most residents uncovered","Fairness validation becomes checkbox theater"]}},"retrievedAt":"2026-09-17T22:17:30-05:00","enrichedAt":"2026-09-17T22:25:00-05:00","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Protects state and municipal workforce members as patients; disclosure rules affect how AI-mediated benefits communications are presented. Does not address accessibility of the disclosure UX itself.","procurementImplications":"Model for AI clauses in health-plan carrier contracts: human review, no-train, disclosure, and audit/governance reporting to a state benefits owner.","operatingModelImplications":"Shifts AI risk ownership to the plan administrator via policy now, statute later; keeps clinical judgment with humans while permitting some automated claims processes.","sourceVerification":{"openedUrl":"https://osc.ct.gov/wp-content/uploads/2026/09/2026.09.16-Comptroller-AI-Healthcare-Policy.pdf","referenceExcerpt":"No adverse determination may be made solely by an AI system, and they must always be reviewed by a human.","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}