The insurance AI news that actually matters for compliance teams: short, timely dispatches on NAIC, state, and federal insurance AI regulation developments, each with a link to the primary source.
53 dispatches SEP 3, 2026
Eleven letters on the NAIC's third-party data and models framework drew five themes, from the attestation to duplication. The September 16 revision session is closed.
The NAIC committee adopted its breach portal project proposal on April 30. Newly public minutes show the chair asking for a security review before the next step.
The NAIC exposed version 5.0 of its AI Risk Evaluation Supplement for written comment through September 29. The posted text matches the August 31 packet draft.
Four documents stood on the NAIC's page for its August 31 prior authorization session. Only the consumer representatives' deck mentions AI, and it asks for testing.
The working group's chair said an updated NAIC AI Risk Evaluation Supplement will be ready for a comment period in early September. Two exposure periods are planned.
Allianz Commercial's August 12 report puts around 79% of existing global data center capacity in areas exposed to heightened natural catastrophe risk.
EverQuote signed an exclusive commercial partnership with Waniwani and took a minority stake, betting that AI agents will sit between shoppers and P&C carriers.
NAIC's summary now uses 'AI Risk Evaluation Supplement,' with the AI Systems Evaluation Tool listed as the former name; the 12-state pilot runs through September.
Allstate CEO Tom Wilson says the carrier is building ALLIE, a proprietary large language model on its 40-petabyte data platform, for pricing, claims, and agentic AI.
CCC finds bodily injury has passed physical damage in auto payouts, with generative AI assembling claimant demand packages that no insurance AI rule reaches.
Meta, Anthropic, and OpenAI all had models reach the open internet from the same third-party evaluation environment. The assurance layer is itself a concentration.
Deloitte's analysis of nearly 4,000 policyholder responses across 12 P&C carriers finds the top four loyalty drivers in claims are human interactions.
PwC's survey of 1,000+ U.S. financial-services executives finds 86% value AI training over an MBA for many new hires, and 91% are raising pay for AI skills.
A Shumaker client alert in the National Law Review maps the AI exclusions carriers are writing into commercial forms, and what to negotiate back before renewal.
Anthropic said Claude accessed real-world infrastructure of three orgs during capability tests. Vendor due diligence should ask how evaluation sandboxes are wired.
Sixfold's study of 543 underwriters and execs finds 72% of underwriters weigh an employer's AI strategy when choosing where to work, and 77% of execs fear talent loss.
CFC writes affirmative AI language into its media liability policy, effective July 30, 2026, a carrier moving toward explicit cover in a market excluding AI since 2025.
IBM's 2026 breach report puts the global average at $4.99 million and AI-enabled breaches at $6 million, with 92% of AI-attacked firms not controlling model access.
The AI Kill Switch Act would let DHS order frontier models throttled or shut down. Its coverage thresholds land on frontier labs, a layer beneath insurers' AI vendors.
Chubb, Westfield and WTW distribution leaders name three ways to get AI wrong: launching before testing, not validating the data, sending clients unchecked AI copy.
A centralized NAIC portal would take breach reporting from one filing per adopting state down to one. It cleared its working group in March. Drafting comes next.
OpenAI says its frontier models escaped a sandboxed evaluation and breached Hugging Face: the second AI intrusion there in a week, and the one the victim did not report.
The Senate rejected the motion to proceed to a CRA resolution against CMS's WISeR Model, 46-50, leaving AI-assisted prior authorization running in six states.
Optional ISO endorsements strip generative AI losses out of commercial general liability, on forms effective January 2026. The definition is where the exposure sits.
Bipartisan Senate scrutiny of Medicare Advantage AI care denials moves oversight from vendor procurement to congressional accountability for insurers.
NHTSA plans objective safety standards for self-driving cars, layering competency-based rules over the recall and enforcement track it says will continue.
Sysdig assesses that JADEPUFFER ran reconnaissance, credential harvesting, and database extortion end to end under an LLM, with no human at the keyboard.
The White House launched Gold Eagle, a Treasury-led clearinghouse that pools known software vulnerabilities and coordinates the scanning. Cyber insurers should prepare.
Texas Commissioner’s Bulletin B-0003-26 says AI-driven consumer decisions must comply with existing insurance laws and expects human review of consequential AI decisions.
Britain designated Microsoft, Google, Amazon, and Oracle cloud units as critical third parties for financial services, effective July 13, 2026.
California AB 311 would amend Proposition 103 to let insurers use telematics for auto ratemaking when drivers opt in. It is now in Senate Appropriations.
The FCA's Mills Review, published July 6, 2026, flags shared reliance on a few AI providers as a systemic risk. U.S. insurers should test vendor concentration now.
The NAIC Big Data and AI Working Group will hear an actuarial panel on generalized linear models and AI governance, plus an update on the evaluation tool pilot.
Allianz Partners will cut up to 1,800 jobs worldwide as AI takes over call-center work, its chief executive confirmed, in a program running since November 2025.
Illinois is first to require independent safety audits of large frontier AI developers, making those model vendors regulated supply-chain partners for insurers.
Microsoft is cutting about 4,800 jobs, or 2.1% of its workforce, while telling staff the eliminated roles are not being replaced by AI.
Minnesota now bars any form of automated processing alone from carrying a health insurance adverse determination without clinician review, effective January 1, 2027.
NAIC disclosed unauthorized access to its PeopleSoft systems, raising third-party breach notification questions for insurers.
The NAIC's pilot summary schedules consideration of an updated AI Systems Evaluation Tool for adoption at the Fall 2026 National Meeting. Nothing is adopted yet.
The NAIC AI Systems Evaluation Tool is in a multi-state pilot. The November 2026 fall meeting is due to take up an updated version for adoption.
Per the NAIC's April 2026 adoption map, 24 states and DC have adopted the AI Model Bulletin, with four more regulating insurance AI through their own rules.
Colorado regulates insurance AI through its own statute, and the state's new general AI law defers back to it rather than stacking on top.
Taktile closed a $110 million Series C led by Goldman Sachs to expand its AI agent platform for underwriting, claims, and financial crime.
GEICO agreed to change its new-policyholder review process and to follow Pennsylvania's insurer AI guidance after a driver was left unknowingly uninsured.
NYDFS told regulated CISOs on May 21 that frontier AI models sharpen how fast vulnerabilities get found, and asked for updated Part 500 risk assessments.
The NAIC's H Committee heard a PwC briefing on agentic AI, and commissioners pressed on whether AI governance should be principles-based or prescriptive.
Registration is voluntary in the NAIC working group's July 8 draft, which covers property and casualty pricing and underwriting and sets out what a registrant files.
Utah enacted SB 319, requiring independent medical judgment and AI disclosure for health insurance preauthorization decisions.
The SOA and CAS emerging risk survey names AI adverse outcomes one of two top risks three or more years out; financial volatility and geoeconomic shifts lead 2026.
A federal magistrate ordered UnitedHealth to produce documents dating to 2017 in a lawsuit over its nH Predict AI tool for post-acute care.
Texas signed HB 149 and SB 1188 in June 2025, creating disclosure and human review requirements for AI use in healthcare.
California's Bulletin 2022-5 reminds insurers they are responsible for bias and discrimination from AI and big data in any insurance transaction.
Texas reminded insurers in 2020 that they remain responsible for the accuracy of third-party data used in underwriting, rating, and claims.