Connecticut’s insurance regulator announced new rules on September 16, 2026 governing how artificial intelligence can be used to deny health insurance claims, but the regulation applies only to the state employee health plan and the Connecticut Partnership Plan, covering roughly 220,000 of the state’s residents. The other three million-plus Connecticut residents covered by private commercial insurance remain outside the new rule’s reach, a gap critics say undercuts the policy’s real-world impact even as it sets an important precedent.
What the New Rule Actually Requires
The regulation requires that AI systems used to evaluate claims for the covered state plans cannot be the sole basis for a denial, a licensed clinician must review and sign off on any adverse determination influenced by an algorithm, and patients must be told when AI played a role in a decision about their care. Connecticut Comptroller Sean Scanlon, whose office oversees the state employee plans, announced the policies and said the rules take effect January 1, 2027. The package also bars insurers from using AI as the sole basis for down-coding claims or cutting a provider’s payment, and it guarantees that member data submitted for claims cannot be reused to train other AI models. The restrictions reach four separate vendors handling different pieces of the state’s coverage: Anthem administers medical benefits, Cigna handles dental claims, Aetna covers Medicare Advantage retirees, and Caremark manages the pharmacy benefit, meaning enforcement depends on each contractor changing internal review workflows on the same timeline. The move follows a familiar pattern in health insurance AI oversight: state and federal regulators moving cautiously and often incrementally, covering government-adjacent plans first before, if ever, extending similar protections to the broader commercial insurance market.
The Lawsuits That Forced the Issue
Connecticut’s action arrives against the backdrop of ongoing federal litigation against major insurers over AI-driven claim denials. A federal class action against UnitedHealthcare alleges the company used an AI model with a 90% error rate to override treating physicians’ determinations and wrongfully deny elderly Medicare Advantage patients care their doctors had ordered; a federal court in Minnesota has allowed key parts of that case to proceed. Separately, Cigna faced scrutiny after a ProPublica investigation revealed the company’s PxDx algorithm rejected more than 300,000 payment requests over a two-month period in 2022, with company doctors reportedly spending an average of just 1.2 seconds reviewing each flagged claim before signing off on the denial.
Two Very Different Views of Whether This Helps
Patient advocacy groups have welcomed Connecticut’s rule as proof that meaningful AI oversight in insurance is achievable, and argue it should be used as a template state legislators nationwide can adapt and expand to cover commercial plans. Insurance industry representatives counter that AI-assisted claims review, when properly implemented, actually speeds up approvals for the vast majority of routine claims and that heavy-handed regulation risks slowing down care for patients whose claims would have been approved quickly anyway, framing the debate as a tradeoff between speed and oversight rather than a simple question of protecting patients from faulty algorithms.
A Regulatory Patchwork Taking Shape
Connecticut is not alone in narrowly targeting AI claims oversight. A companion measure, HB5587, has separately moved through the state legislature to prohibit health insurers from using artificial intelligence in certain claims decisions, running parallel to Scanlon’s administrative rule for the state employee plans. Other states have introduced or passed similar measures covering specific populations, and courts have separately begun allowing broader discovery into how insurers actually deploy AI models internally, litigation that could eventually reveal error rates and override patterns regulators have struggled to obtain through public rulemaking alone. The result so far is a state-by-state patchwork where protections often apply to whichever population a legislature or regulator had the political capital to cover first, typically public employees, rather than a comprehensive national standard.
The Comptroller’s Case and the Push for a Broader Law
Scanlon framed the rule as a first step rather than a finished policy, telling reporters when the regulations were announced, “Our world is changing rapidly, and government needs to change with it at the same speed in order to protect people.” His office says it intends to ask the General Assembly in its January 2027 legislative session to extend the same protections, including the ban on AI-only denials and the clinician sign-off requirement, to every state-regulated health plan in Connecticut, not just the roughly 220,000 people on state employee and Partnership Plan coverage. Whether that broader bill succeeds will determine if Connecticut’s approach becomes a statewide standard or remains limited to public-sector workers, since Scanlon’s own regulatory authority as comptroller only extends to the plans his office administers and cannot on its own reach commercial insurers or the fully insured market.
What to Watch For Next
The test for Connecticut’s rule will be whether it demonstrably reduces improper denials for the populations it covers between now and its January 1, 2027 effective date, evidence state lawmakers elsewhere are likely to demand before extending similar protections to commercial insurance. Meanwhile, the active federal lawsuits against UnitedHealthcare and Cigna are working their way through discovery and motions to dismiss, and a ruling favorable to plaintiffs in either case could do more to reshape industry-wide AI claims practices than incremental state regulation, simply by exposing internal error rates and documentation insurers have so far kept confidential. Patient advocates in Connecticut say they will be watching implementation closely over the next year to see whether the clinician-review requirement is applied meaningfully, with real time spent on each flagged case, or becomes a rubber-stamp formality similar to what plaintiffs allege happened inside Cigna’s PxDx review process, where a doctor’s signature was attached to a denial without substantive independent review of the underlying claim.
Photo: SHVETS production / PEXELS via Pexels
