Washington SB 5395 creates the first state-level disclosure regime in the country requiring health carriers to report, annually to the insurance commissioner, the percentage of prior-authorization denials aided by AI tools and algorithms (Washington Legislature, March 2026). The reporting section anchors a six-state patchwork of 2026 laws restricting AI in adverse coverage determinations, with Iowa HF 2635 and Indiana HB 1271 both effective July 1, 2026.

6
States With 2026 AI Health Laws
Jul 1
Iowa & Indiana Effective Date
Jun 11
Washington SB 5395 Effective
25
NAIC Bulletin Adoptions

Key Takeaways

  • Iowa HF 2635 and Indiana HB 1271 both took effect July 1, 2026, inside a six-state slate of AI-restrictive health insurance statutes running through January 2027.
  • Washington requires carriers to report "the percentage of total denials that were aided by artificial intelligence tools and algorithms," disaggregated to health plan and delegated benefit manager.
  • From 10.0% in 2020 to 22.7% in 2022 is how one large carrier's Medicare Advantage post-acute prior-authorization denial rate moved across its automation build-out.
  • 80.7 percent of appealed Medicare Advantage prior-authorization denials are overturned, which puts substantial noise into any initial denial signal regardless of AI involvement.
  • Twenty-nine jurisdictions now operate a formal AI insurance regime, 25 through the NAIC bulletin plus four with their own, at a different regulatory layer from these six statutes.

The Six-State Slate and the July 1 Dates

Six states enacted AI-restrictive health insurance statutes taking effect across 2026 and January 2027: Alabama, Georgia, Indiana, Iowa, Utah and Washington (National Law Review, July 2026). Maryland HB 1563, effective June 1, adds a seventh with quarterly reporting to the state commissioner. Each addresses the same concern from a different structural angle: whether an algorithmic system can be the terminal decision-maker on a coverage denial, delay or downcode.

StateBillEffectiveCentral Provision
MarylandHB 1563Jun 1, 2026Quarterly reporting of adverse decisions and AI involvement
WashingtonSB 5395Jun 11, 2026Licensed physician for denials; annual AI-share reporting
IowaHF 2635Jul 1, 2026URO framing; AI not sole basis for deny/delay/downgrade
IndianaHB 1271 (PL 88)Jul 1, 2026No AI as sole basis for claim downcoding; 180-day recoupment cap
AlabamaSB 63Oct 1, 2026Individual-basis PA decisions; annual anti-discrimination certification
UtahSB 319Jan 1, 2027Disclosure to DOI, providers, enrollees; no sole AI reliance
GeorgiaSB 544Jan 1, 2027Licensed provider review before adverse determination

Washington SB 5395 was signed March 25 and takes effect June 11. It bars carriers, healthcare benefit managers and public employee health plans from relying exclusively on AI to deny prior authorization, requires a licensed physician or other licensed professional to make the adverse determination, and layers the annual reporting obligation on top.

The two July 1 statutes regulate different parties. Iowa HF 2635 permits utilization review organizations to use AI for initial review but forbids it as the sole basis for denying, delaying or downgrading on medical necessity, with denials made by a qualified reviewer or clinical peer. The URO framing reaches the vendors carriers outsource prior-authorization workflows to, not only the carriers. Indiana HB 1271, now Public Law 88, restricts downcoding specifically and requires adverse determinations to disclose AI use, alongside a companion provision capping overpayment recoupments at 180 days from initial payment.

Alabama SB 63, effective October 1, requires AI-assisted decisions to rest on the individual beneficiary's medical history and circumstances, with annual certification that systems do not discriminate and undergo accuracy monitoring (Holland & Knight, May 2026). Utah SB 319 and Georgia SB 544 take effect January 1, 2027.

The First Benchmark Dataset, and What It Costs to Produce

Carriers writing at least one percent of Washington accident and health premium must report total prior-authorization requests, approvals and denials; the breakdown by health plan and each delegated benefit manager; response-time metrics; and the percentage of total denials aided by AI. The commissioner aggregates it into a public report without identifying carriers, with AI-transparency reporting beginning January 2027.

That structure produces a numerator against a denominator at carrier level for a variable previously either aggregated nationally or reachable only through litigation discovery. One large national carrier's Medicare Advantage post-acute prior-authorization denial rate rose from 10.0% in 2020 to 22.7% in 2022 across the period it was implementing automation (Nature Digital Medicine, June 2026). Washington will make cross-carrier variance of that kind observable in ordinary regulatory filings.

The compliance cost lands before the data does, and it lands as a case-mix event in medical-management economics rather than as a governance line item. Indiana's downcoding restriction moves loss adjustment expense from algorithm-only pipelines back to case-manager review for the subset of claims where downcoding is contemplated. Unit cost of downcoding review rises; volume of downcoded claims may fall as the higher unit cost prices marginal cases out of the workflow. Both effects run into 2027 loss ratios and administrative expense loads.

Iowa moves the same cost upstream. A carrier contracting a third-party utilization review vendor has to confirm the vendor's AI workflow satisfies HF 2635, and if it does not, either the vendor reworks its process or the carrier absorbs the change through renegotiated terms. The vendor economics supporting low-cost outsourced prior-authorization review shift either way.

Neither effect appears in prior-year experience data. A 2027 filing in any affected state is projecting a workflow that did not exist in the base period, and state DOI actuaries reviewing those filings now have a statutory hook for asking whether the operational model satisfies the applicable statute. Filings projecting reductions in medical-management expense from AI adoption should expect exactly that question.

Two Compliance Tracks and a Confounded Denominator

The statutes sit alongside the NAIC Model Bulletin on the Use of Artificial Intelligence Systems, adopted December 2023 and now adopted or referenced in 25 states plus the District of Columbia, with four further jurisdictions running their own frameworks for a total of twenty-nine.

The two operate at different layers, which is why satisfying one does not satisfy the other. The bulletin requires a written AI Systems Program with senior-management and board accountability, risk controls, model validation and testing for bias, and third-party oversight, across all lines and use cases. The six statutes are line-specific and use-case-specific: they regulate which decisions the AI may make and what must be reported to the commissioner. A multistate carrier extends the first with the second, on top of the preexisting state AI patchwork and Colorado's own insurance-specific regime.

The dataset the reporting section produces is also harder to read than its structure suggests. Case mix differs between AI-triaged and human-only queues, the routing rules that assign a request to one queue or the other are the carrier's own, and appeal-overturn dynamics may differ across the two paths. Nationally, 80.7 percent of appealed Medicare Advantage prior-authorization denials are overturned, which puts substantial noise into the initial denial signal whatever produced it.

So the AI-assisted share benchmarked against the observed denial rate is a rough proxy for whether AI-assisted denials cluster at higher or lower rates than human-only decisions inside the same carrier, and not more than that. What it does supply is a distribution rather than a point estimate, computed with the state-level regulatory environment held constant, which is an external anchor no credibility-weighted denial analysis has had before.

The pressure is converging from the federal side too. CMS's prior-authorization interoperability rule already required payers to publish annual approval rates, denial rates and average decision times, with first reports due March 31, 2026, and CMS has been signalling that explainable AI is the direction of travel for Medicare and Medicare Advantage. Between a federal explainability track and state statutes requiring a licensed human decision-maker, a black-box scoring model applied to adverse determinations is squeezed from both sides at once, and the identity of the terminal decision-maker has become a regulated feature of the product rather than an operational choice behind it.

Further Reading

Sources