State & Local GovernmentIssue 503 records
State Government
No new measured state AI results today. Instead, three items show states working out how to keep AI answers tied to official information. Oregon published a detailed disclosure card for its Medicaid work-rules chatbot: information only, personal data stripped, chats kept a year, retested when policy changes. Ohio's secretary of state asked voters to report wrong election answers from ChatGPT, Claude and Grok. And an independent study finds those platforms now make few factual errors but send voters to state election sites less than half the time.
- Evidence records
- 3
- Cross-source patterns
- 1
- Evidence classes
- 1 government evaluation1 independent reporting1 independent research
- Outcomes
- 2 emerging1 mixed
- Source freshness
- 2 recent1 new this fortnight
- Research completed
- 2026-10-29
Choose a role to see its takeaway beside every record in the ledger.
Patterns across the evidence
Synthesis · Lighthouse Advisory interpretation
States are treating their official information as the yardstick for AI answers, checking their own chatbots and outside platforms against it
Interpretation. Oregon tests its Medicaid chatbot by having experts compare answers with approved information and retests when policy changes. Ohio asks voters to report AI answers that conflict with official election sources. The States United study finds the main weakness of commercial AI is not factual errors but failing to point voters to state sites. Together they suggest that keeping official content current, structured and easy to cite is becoming core AI governance work for agencies, whether the AI is the state's own or someone else's. None of the sources yet reports whether these checks reduce errors.
Operating questionWhich official source is the reference for each AI-answered topic, who keeps it current, and how often are AI answers, both the state's and major platforms', tested against it?
Supporting evidenceOregon Department of Human Services (ODHS); Oregon Health Authority (OHA); Oregon Enterprise Information Services (EIS)Ohio Secretary of StateStates United Democracy Center
Evidence ledger
Full record · every source keeps its link and limitations
Oregon publishes a disclosure card for its generative AI chatbot on Medicaid work rules: information only, no eligibility decisions, PII stripped before the model, chats kept one year, retested on policy or model changes
ODHS and OHA run a generative AI chatbot that answers questions about federal Medicaid work or activity rules for the Oregon Health Plan, which take effect Jan. 1, 2027. The card states the chatbot provides information only and does not decide eligibility or coverage. Deloitte built and supports it under contract; it uses Anthropic models including Claude Sonnet. It is designed to use only Level 1 or Level 2 information, to detect and remove personal or health information before it reaches the model, and to keep data in the United States; conversations are not used for model training. Prompts, answers and session IDs are kept for one year and may be public records. Testing covers accuracy against approved information, fairness, privacy, security and resource use, and is repeated after major policy, model or instruction changes or a rise in reported problems. The chatbot is English-only and can be turned off at any time. EIS requires such cards for public-facing AI chatbots.
Why it matters, evidence and limitations
- Why it matters
- Why now: Oregon's Chief AI Officer discussed this chatbot and its disclosure card in a Lawfare Scaling Laws episode released Oct 27, and Medicaid work rules start in about two months, so states are standing up resident-facing AI help on a tight timeline. Not in our archive. It is a rare published, state-mandated description of a live benefits chatbot's data limits, retention, testing triggers and vendor, which other states can compare against their own.
- Evidence and measured results
- Official agency disclosure card opened in full; Lawfare episode page opened (description only; audio not reviewed). The card describes design and testing categories but reports no accuracy rates, usage volumes, launch date or cost.
- Limitations and uncertainty
- Self-description by the agency, not an independent evaluation; no performance metrics published. English-only at time of review. Podcast audio not reviewed, so Chief AI Officer statements are not cited.
Ohio's secretary of state asks voters to send screenshots of wrong or misleading election answers from ChatGPT, Claude, Grok and other AI platforms, a week before the Nov. 3 election
Secretary of State Frank LaRose announced an effort to ensure voters get accurate election information from AI platforms including ChatGPT, Claude and Grok. The office says it works with technology partners and urges voters to verify dates, hours, polling places, ID rules and deadlines with VoteOhio.gov, the secretary of state or a county board of elections. Voters who receive false or misleading Ohio election information from an AI platform are asked to send screenshots showing the prompt, the complete response and the platform, by email or to @VerifyOhio on X, Facebook or Instagram.
Why it matters, evidence and limitations
- Why it matters
- Why now: announced Oct 27 and not in our archive. It is a state agency treating commercial AI answers as part of its information environment and setting up a reporting channel for errors, rather than only governing AI it runs itself. It follows our Oct 26 item on Washington's proposed AI signature checks; election offices are now dealing with AI on both sides.
- Evidence and measured results
- One local news report opened in full, closely following the office's announcement with direct quotes. The secretary of state's own site returned a bot-check page and was not opened. No figures on reports received, partner names or how corrections will be sought.
- Limitations and uncertainty
- Announcement only; no data on volume, error types or platform responses. Single news source. Timing one week before the election limits what can be corrected before voting ends.
States United study: ChatGPT and Google AI cut factual errors on voter questions to 0% in testing, but pointed voters to state election websites in under half of answers and often gave incomplete candidate lists
States United tested ChatGPT and Google AI on common voter questions in two rounds: 497 responses across six states in Sept-Oct 2025 and 402 responses across Arizona, Michigan and Pennsylvania in Jan-Feb 2026. Verifiable factual error rates fell from 6.9% (Google AI) and 8.2% (ChatGPT) to 0% in the second round, but answers were often incomplete. ChatGPT mentioned a state election site in 39.4% of responses and Google AI in 55.6%. ChatGPT returned incomplete candidate lists for 88.9% of gubernatorial queries. Of 3,481 cited links, state and local government sources made up 41.5% and Wikipedia 12.3%. Google AI switched to links-only answers around Feb. 2, 2026. Recommendations for election officials include structuring official sites for AI discoverability and building voter education around candidate filing windows.
Why it matters, evidence and limitations
- Why it matters
- Why now: published in August but newly relevant because Ohio's secretary of state launched an AI error-reporting channel on Oct 27, a week before the Nov. 3 election. It is the independent evidence check on that approach: platforms have become more accurate on simple facts, and the bigger gap is that they often do not send voters to official state sources. Not in our archive.
- Evidence and measured results
- Full report page opened, including findings, recommendations and methodology: standardized questions coded by researchers across two rounds and three platform conditions. Limits: two platforms, a few states, testing before the general-election period; not peer reviewed.
- Limitations and uncertainty
- Advocacy research organization; small state sample; platform behavior changes quickly, as the Google format change shows. Results may not hold for Claude, Grok or other platforms named by Ohio.
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.
- Government evaluation
- A public body’s measured evaluation or documented pilot.
- Independent reporting
- Independent reporting with attributable sources but without a formal evaluation design.
- Independent research
- Research conducted outside the implementing organization.