Health & Human ServicesIssue 044 records
Health & Human Services
Four sources, mostly about AI controls. Michigan's health and human services department put an AI policy into effect Nov. 1. It requires approval before vendor AI uses department data, expert review of AI output, and a non-AI option for the public. New York's Medicaid eligibility system RFP (Oct. 29) asks the winning vendor to use AI but requires 90 days' notice and written approval, with non-essential AI off by default. The Justice Department alleges a Minnesota disability-services provider used AI to fake patient notes for $3.2 million in Medicaid claims. WHO/Europe's 53 countries adopted an AI-in-health roadmap. No new evidence on public hospitals, child welfare or aging this edition.
- Evidence records
- 4
- Cross-source patterns
- 2
- Evidence classes
- 3 standards or public-body guidance1 government audit
- Outcomes
- 3 emerging1 cautionary
- Source freshness
- 4 new this fortnight
- Research completed
- 2026-11-03
Choose a role to see its takeaway beside every record in the ledger.
Patterns across the evidence
Synthesis · Lighthouse Advisory interpretation
HHS agencies are writing AI approval gates into policy and contracts
Michigan's HHS department now requires dual governance approval before any vendor AI uses its data, plus expert review before outputs are acted on. New York's Medicaid eligibility RFP encourages vendor AI but requires 90 days' notice, written approval, disabled-by-default settings and no training on department data. One is an internal policy and the other a contract, but both require approval before use and let the agency keep the right to refuse.
Operating questionFor each AI use touching eligibility or case data, who approved it, under which policy or contract clause, and how would you switch it off?
Supporting evidenceMichigan Department of Health and Human Services (MDHHS)New York State Department of Health (Office of Health Insurance Programs)
AI-produced records need provenance, not just review
DOJ alleges a Minnesota provider used AI to fabricate patient notes supporting Medicaid claims. Michigan's new policy requires staff to cite AI contributions and warns that AI transcripts may be FOIA-disclosable. New York's RFP turns AI transcription off by default and gives the department ownership of prompts and outputs. Taken together (interpretation), agencies are starting to treat who or what produced a record as something that must be traceable.
Operating questionCan you tell which case notes, transcripts or claim documents were drafted by AI, and would your audits catch a well-written record of a service that never happened?
Supporting evidenceU.S. Department of Justice (Fraud Division and U.S. Attorney's Office, District of Minnesota)Michigan Department of Health and Human Services (MDHHS)New York State Department of Health (Office of Health Insurance Programs)
Evidence ledger
Full record · every source keeps its link and limitations
Michigan MDHHS AI acceptable-use policy takes effect Nov. 1: approved tools only, expert review of every AI output, and a non-AI alternative for the public
MDHHS policy APS 1100-02, issued Oct. 22 through bulletin APB 2026-041 and effective Nov. 1, 2026, sets department-wide rules for AI. It covers employees, contractors, temporary workers and third parties, and all AI developed, procured or deployed by DTMB or MDHHS vendors. Rules: AI only on state-managed devices. No internal, confidential or restricted data in external AI systems. Vendor or software AI may use MDHHS data only for use cases approved by both the DTMB AI Governance Board and the MDHHS Data Governance Team. Sharing MDHHS data with an unapproved tool counts as an unauthorized disclosure, reportable on form DCH 1422; policy violations must be reported within 24 hours of discovery. A subject matter expert must review AI output before anyone acts on it. AI systems need human-in-the-loop. AI contributions must be cited. Chatbots must say they are AI and offer the public a non-AI alternative. AI recordings and transcripts may be subject to FOIA. Next review is set for Nov. 1, 2027.
Why it matters, evidence and limitations
- Why it matters
- Medicaid, SNAP, child welfare and public health all sit in one department, so this one policy governs AI across a large state HHS agency from the day it takes effect. The ACLU of Michigan has already questioned MDHHS's use of AI in SNAP payment assessments (archived in this stream, 2026-10-29). The policy shows the controls now on paper; it does not show how they are enforced.
- Evidence and measured results
- Primary policy text (6 pages) plus the 2-page bulletin announcing it. It sets rules and reports no outcomes, metrics or inventory of approved tools.
- Limitations and uncertainty
- A policy is not evidence of compliance or of results. The source does not list approved tools, how the governance bodies decide, or how SME review will be audited. Whether the policy reaches already-deployed eligibility tools is not stated. The policy applies to MDHHS staff and vendors, not to counties or providers outside its scope.
New York DOH's seven-year Medicaid eligibility system RFP asks vendors to use AI, but every use needs 90 days' notice and written approval, with non-essential AI off by default
The RFP, issued Oct. 29, 2026, seeks one contractor for a seven-year term. The contractor would maintain and operate New York's Medicaid Eligibility and Client Management (MECM) system and Appeals Management System. It would also fold NY State of Health functions into MECM, creating one Converged E&E System for Medicaid, Child Health Plus, the Basic Health Program and Qualified Health Plans. Section 5.2.9 encourages the contractor to use AI to improve quality and cut time and cost. The department's vision cites AI-enabled, model-driven solution engineering to lower costs. Section 6.10 controls that AI. The contractor must give 90 days' written notice and get written approval before using any AI. AI not integral to the system is disabled by default. The department may disallow any AI for any reason, and rejected AI must be replaced at no cost. Department data and AI outputs may not be used to train models. The department owns prompts and outputs. AI transcription is off unless requested. The department may ask for an independent audit or a NIST-defined AI impact assessment. Letters of intent are due Nov. 13, 2026; proposals Feb. 19, 2027.
Why it matters, evidence and limitations
- Why it matters
- The RFP covers 6.5 million enrollees across the programs (Table 2, July 31, 2026), so it is one of the largest state eligibility procurements in the country. It shows how a state Medicaid agency can encourage vendor AI for engineering productivity while keeping tight approval and data controls. Other states can reuse the clause set in their own eligibility contracts.
- Evidence and measured results
- Primary procurement document. It sets requirements and states expected outcomes; it reports no measured results from AI use.
- Limitations and uncertainty
- This is a solicitation, not an award or a deployed system. Requirements may change through Q&A (responses due on or about Jan. 19, 2027) and Attachment E, the requirements matrix, which was not reviewed. Expected AI cost savings are department aspirations. The 6.5 million total is NYSOH enrollment as of July 31, 2026, not the full Medicaid caseload.
DOJ alleges a Minnesota Medicaid disability-services provider used AI to create fake patient notes backing $3.2 million in claims
DOJ's Oct. 30 release lists 13 Minnesota Medicaid and SNAP fraud defendants charged, convicted or sentenced since Sept. 30. It puts intended losses at nearly $83 million, about $29.1 million of it in newly charged cases. One defendant, a 68-year-old Maple Grove man, was indicted Oct. 28 on eight counts of health care fraud. The charges cover about $3.2 million in claims to Minnesota Medicaid's Integrated Community Supports program, a disability-services program. Prosecutors allege he billed for services not provided, including home services while recipients were traveling, services while providers were working other jobs, and services recipients never received. They also allege he used 'Super Intelligence' to create fake patient notes to justify the claims. The release does not name or describe the tool, and the charges are allegations only.
Why it matters, evidence and limitations
- Why it matters
- State Medicaid agencies and Medicaid Fraud Control Units often check service notes before or after payment. This is a federal allegation that AI-generated notes were used to support false home and community-based services claims. If true, documentation that looks complete is weaker proof that a service was delivered. That matters for HCBS reviews, electronic visit verification and audit sampling.
- Evidence and measured results
- Government press release summarizing an indictment. No court findings; the release gives no detail on the tool, the number of notes or how they were detected.
- Limitations and uncertainty
- These are allegations in a single indictment, and the defendant is presumed innocent. The release spells his surname two ways. It does not say what 'Super Intelligence' refers to. Federal agencies also use the term for AI generally, so it may not be a product name. This is one case in one state; it shows a risk, not how often it happens. The evidence class was set to government-audit as the closest available label for a federal enforcement announcement; it is not an audit.
53 WHO European Region countries adopt first AI-in-health roadmap; WHO/Europe says only 1 in 4 countries trains working staff in AI
At the 76th Regional Committee, which closed Oct. 29, 2026, the 53 Member States adopted the Region's first 'Roadmap for action on artificial intelligence in health' (2026–2033). They also unanimously adopted a healthy-ageing strategy, 'Ageing is living'. The release cites a recent WHO/Europe readiness report. Nearly 2 in 3 countries use AI for diagnosis and half use conversational AI. Only 1 in 5 teaches AI to health students and 1 in 4 trains staff already at work. Fewer than half have checked whether their laws cover AI, almost 40% have no ethical guidance, and 8% have a health-specific AI strategy. WHO/Europe says it wants proof of safe and effective care before any tool is scaled up. A midterm report in 2029 will assess the roadmap.
Why it matters, evidence and limitations
- Why it matters
- Limited and indirect for U.S. SLED. U.S. state and local public health agencies and public hospitals face the same gaps: AI is in use ahead of rules, workforce training is thin, and accountability is unclear. The readiness figures give a reference point for health-agency governance work. Transfer to U.S. agencies is interpretation.
- Evidence and measured results
- News release summarizing an adopted roadmap and figures from a separate WHO/Europe readiness report. The roadmap and report were not opened; the figures are as stated in the release.
- Limitations and uncertainty
- Regional intergovernmental policy, not U.S. evidence. The roadmap text and the readiness report behind the percentages were not reviewed; survey methods and response rates are not given in the release. The roadmap is non-binding even for Member States.
How to read this edition
Source findings, measured results and limitations come from the cited publications. Patterns, operating questions, role takeaways and implementation considerations are Lighthouse Advisory interpretation, stated as questions to validate locally rather than guaranteed outcomes. Vendor and operator claims are labeled as claims. Full research method.
- Standards or public-body guidance
- Normative or advisory guidance from a standards body or public institution.
- Government audit
- An oversight review of performance, controls, or operations.