Signal Scan · ARTIFICIAL INTELLIGENCE & AUTOMATION · 7 November 2026

Insurers Write AI Out of Liability Cover While Supervisors Examine Insurers’ Own AI Use

Liability insurers are writing generative AI out of commercial policies through state-reviewed endorsements, while insurance supervisors in the US and EU build tools to examine insurers’ own AI use, leaving the terms on which firms can deploy AI to policy wording through 2028.

Produced using Shaping Tomorrow’s six-stage process for converting external change into decision-ready intelligence.
Scanned and filtered by AI. Stress-tested and judged by humans.

The consensus reading is that AI is governed by statute and supervisory guidance. Beneath it, a quieter rulebook is being written in policy wording. Carriers are filing endorsements that strip generative AI from commercial liability cover, and CSIS, citing press reporting, says state commissioners have approved more than 80% of such requests (CSIS, 04/09/2026). The supervisors who review those filings are meanwhile building tools to examine how insurers use AI themselves (Alston & Bird, 21/09/2026). The sources show no public body setting the terms on which an AI deployer can be insured. The inflection arrives at the 2027 renewal cycle, when exclusions meet claims.

Signal Identification

A regulatory pivot by default: whether AI deployment is insurable is being decided by policy wording, reviewed filing by filing, not by a published standard. Insurance supervisors aim their AI work at insurers’ internal governance. The gap between those two tracks is the signal.

Time horizon: 1-3 years, and it binds 2027 to 2028, shown when a carrier’s refusal of an AI-related loss under an AI exclusion appears in a court filing, regulatory filing or company disclosure. (Milestones: NAIC Big Data and AI Working Group meets on 16 November 2026; first renewals carrying the exclusions follow in 2027.)
binds 2027 to 2028202720282029
Probability band: Medium-High (60 to 80%)
Probability band: Medium-High, 60 to 80%, the chance the signal binds within the window shown above.LowMediumHigh
Geographic / Jurisdictional Scope: Primary: US states, through state insurance regulation and the NAIC. Spillover: EU, through EIOPA, and the London specialty market.
PrimaryUS states (NAIC)
SpilloverEU (EIOPA)London market
Sectors exposed:
Commercial liability insurers and reinsurersEnterprises deploying generative AIAI model developersInsurance brokers and risk advisersLegal, risk and procurement functions

What's Changing

Insurers are filing, not waiting. The Insurance Services Office has circulated three generative AI exclusion endorsements for general liability (CG 40 47, CG 40 48 and CG 35 08), and carriers are submitting filings to use them with state regulators at a growing rate (Cloud Security Alliance, 30/08/2026). CSIS lists Berkshire Hathaway, Chubb, Travelers and AIG among the filers, and notes that AIG has said it has no current plans to use the exclusion but has secured the option (CSIS, 04/09/2026). Exclusions are also migrating: Berkley’s absolute AI exclusion spans D&O, E&O and fiduciary products, while Beazley and QBE set AI sublimits near 10% of cyber limits rather than excluding AI (Cloud Security Alliance, 30/08/2026).

The cover that remains is thin. CSIS reproduces a Geneva Association matrix in which 13 of 112 cells record AI cover as available and 9 as limited; the remaining 90 are excluded (CSIS, 04/09/2026). At the top of the chain, OpenAI has reportedly secured cover of up to $300 million, while Anthropic agreed to pay a $1.5 billion settlement from its balance sheet (CSIS, 04/09/2026).

Supervisors are looking elsewhere. Twelve state insurance departments are piloting the NAIC’s AI Risk Evaluation Supplement to assess insurer AI governance, with adoption to be considered at the Fall National Meeting (Alston & Bird, 21/09/2026); Coverager calls it “a structured way to ask how an insurer uses and oversees AI” (Coverager, 13/08/2026). The working group’s charges cover regulators’ review of AI systems used by licensees (NAIC, accessed 07/11/2026).

Most AI risk types are excluded across the principal liability lines, and cover for the labs trails their claims

AI risk by policy line: verdicts 112 cells 90 excluded 9 limited 13 available 14 AI risk categories against 8 lines of cover Cover set against a claim OpenAI cover up to $300m Anthropic $1.5bn paid 5 times apart Axis: $0 to $1.6bn Different firms, shown for scale

Source basis: CSIS, 04/09/2026, reproducing a Geneva Association matrix (October 2025) and Financial Times reporting on OpenAI cover; the two dollar figures concern different firms and different kinds of exposure.

Disruption Pathway

Three stages. Through 2027, adoption: exclusions reach general liability, D&O and E&O at renewal, where policyholder negotiation decides how many carriers use them (Insurance Journal, 17/08/2026). Through 2028, testing: claims arise where AI is alleged rather than proven, since Berkley’s wording reaches “any actual or alleged use” of AI (CSIS, 04/09/2026), and courts read the forms. From 2028, sorting: affirmative cover from specialist carriers prices the firms that can show controls, as underwriters converge on a documented human override, an inventory of consequential AI decisions and a named accountable executive (Cloud Security Alliance, 30/08/2026).

Stress concentrates on three points: risk managers who assume silent cover still responds, procurement teams that accept supplier AI without checking supplier cover, and AI developers whose cover trails their exposure. Two adaptations follow. Operationally, firms inventory AI uses against policy wording before each renewal. Financially, boards set an explicit retention for AI loss.

Why This Matters Now

Boards, general counsel, CFOs and chief risk officers hold AI deployments approved on the assumption that liability sits inside the insurance tower. Our reading is that exclusion wording, more than statute, will set the practical limit on deployment in 2027, since CSIS notes that for most businesses an activity that cannot be insured is effectively prohibited (CSIS, 04/09/2026). The decision architecture that needs revising is the AI approval gate: it should record what each deployment does to the firm’s insured position and who holds the retained loss. The NAIC tools examine insurers’ own AI; nothing in this evidence base points to a standard for exclusion wording.

Decision-action posture for this signal: Prepare: policy wording can be audited now, but how widely carriers will use the exclusions is unknown until renewals, so build the inventory and commit retention decisions on a named trigger.

Counter-Argument

The strongest objection: exclusions may stay mostly on paper. AIG holds the filing but has no current plans to use it (CSIS, 04/09/2026). Insurance Journal’s lawyer expects some carriers to price AI risk for additional premium instead of excluding it, and says how widely the forms are used may not be known until renewals start to come in (Insurance Journal, 17/08/2026). Beazley and QBE sublimit rather than exclude (Cloud Security Alliance, 30/08/2026), and affirmative cover from specialist carriers exists.

Partly right, and the objection supports the signal. Even where cover is priced rather than excluded, terms are set carrier by carrier with no public standard behind them; CSIS quotes Lockton Re saying CGL insurers do not currently model, underwrite or price AI risks (CSIS, 04/09/2026). On the available evidence, a sublimit near 10% of cyber limits still leaves most of a large AI loss with the policyholder.

Implications

Taken together, the sources suggest durable change: exclusion wording is the first AI rule many deployers will meet, set without public criteria. The window runs from the 2027 renewals to the first contested claims in 2028. EIOPA’s June report, the canonical supervisory statement on the sector, treats AI for insurers as a cyber and operational-resilience matter (EIOPA, 24/06/2026), not a coverage question. Specialist carriers that can verify controls gain; deployers with undocumented AI use and developers with capped cover lose.

Early Indicators to Monitor

Disconfirming Signals

Strategic Questions

Keywords

AI liability insurance; generative AI exclusions; silent AI; ISO CG 40 47; commercial general liability; D&O and E&O cover; NAIC; AI Risk Evaluation Supplement; insurability of AI; state insurance regulation; EIOPA; AI risk retention

Bibliography

Source tiers: Tier 1, governments, regulators and intergovernmental bodies. Tier 2, think-tanks, academic institutes, major consultancies and quality data providers. Tier 3, quality journalism and specialist trade press. Tier 4, vendor, company and practitioner sources, used only as directional corroboration.

How this was made

The Decision-Grade Intelligence Method

Shaping Tomorrow’s six-stage process for converting external change into decision-ready intelligence.

Scan · Filter · Stress-test · Judge · Trigger · Actionable Intelligence

See full methodology → · How we verify → · Track record →

This Signal Scan is one component of a Shaping Tomorrow intelligence cycle on Artificial Intelligence & Automation. Every monitored topic also includes monthly Change Trackers that assess what is materially shifting in momentum, and a quarterly Decision Intelligence briefing that converts the cycle into board-level strategic implications. Cycles may also close with Action Triggers, an end-of-cycle synthesis that reads across the three formats and surfaces patterns no single deliverable carries on its own. Continuous intelligence on this topic is available as a Single-Topic or Multi-Topic programme.

Discuss a Tailored Programme

Prepared by Shaping Tomorrow: 7 November 2026