Insurance Gaps Mean Tech Acquisitions Can Inherit AI Liabilities

This Insight was published as a Bloomberg Law Perspective in October 2026.

The Bottom Line

A critical issue for acquisitions of technology firms is that a reduction of insurance coverage for artificial intelligence-related losses is often invisible.
The typical insurance operating agreement, which restricts its target from materially altering its policy, may not prevent it from accepting a renewal quotation with an AI exclusion not contained in the prior policy.
From the target’s perspective, nothing improper happened. It maintained continuous coverage on the terms its carrier offered. For the acquiring company, however, the insurance coverage it diligences at signing may not be the coverage it acquires.

The insurance representation in your last technology acquisition was probably accurate. It also may have been worthless. Buyers of AI-enabled companies may be inheriting liabilities they incorrectly believe are insured, and representation and warranty insurance fails to fill the gap.

The usual formulation warrants that the target’s policies are in full force and effect, that premiums are paid, and that no notice of cancellation has been received. A target company whose general liability, directors and officers, and errors and omissions insurance coverage were endorsed at renewal to exclude generative artificial intelligence losses satisfies every word of that representation. But they won’t cover the exposure the buyer cares about most.

That gap can open quietly during the year because the change may appear in documents no one requests. Insurers have moved decisively to eliminate so-called silent AI coverage, but deal practice has not caught up. A buyer may therefore diligence a target’s insurance coverage at signing without discovering that a subsequent renewal has materially narrowed that coverage.

What Changed

The Insurance Services Office, a Verisk unit whose standard forms underpin much of the U.S. property and casualty insurance market, has issued three endorsements addressing generative artificial intelligence. An insurance endorsement is a provision that changes the coverage provided by a standard policy.

CG 40 47 is the broadest of the three. It excludes coverage for bodily injury, property damage, and personal and advertising injury arising out of generative AI under a commercial general liability policy. CG 40 48 is narrower, excluding only personal and advertising injury arising out of generative AI. CG 35 08 addresses a different part of the policy, excluding coverage for bodily injury and property damage arising out of generative AI in connection with products and completed operations. All three forms carry a January 2026 edition date.

The reach of the forms matters as much as the wording. ISO drafts standard language, files it with state insurance departments and licenses it to member insurers. That is why policies from different insurers often contain nearly identical language. When ISO releases a new exclusion, that language becomes available for insurers across the commercial market to incorporate into their policies.

Carriers have also developed their own exclusions. W.R. Berkley’s “absolute” exclusion of AI from its coverage reaches directors and officers, errors and omissions, and fiduciary liability in one endorsement. It sweeps in not only the use and development of AI but statements a company makes about its AI strategy.

The critical issue for deals is that the retrenchment is often invisible. As Fenwick has documented, coverage is narrowing not only through conspicuous exclusions but through revised base forms, altered definitions, new application questions, underwriting file positions and trimmed exceptions to coverage exclusions. A policy holder can renew in good faith, believe nothing changed… and be wrong.

Points of Failure

The problem is not necessarily a failure of diligence. It is that ordinary acquisition diligence is built to confirm that insurance exists, not to uncover how its coverage may have changed. Several routine steps in the process can therefore allow an AI exclusion to pass unnoticed.

The document request list. It typically calls for policies, certificates of insurance and loss runs. It doesn’t call for the endorsement schedule, which is where an attached AI exclusion actually appears. A target can produce a complete, responsive set of documents without the exclusion ever surfacing. Certificates are worse than useless here because they summarize limits and lines without reciting endorsements.

The representation. “Full force and effect” is a status warranty, not a scope warranty. It tells the buyer the policy exists, not whether it covers the target’s principal operating risk. When a target uses AI in hiring, customer communications, underwriting, pricing, or its products, the customary representation measures the wrong thing.

The applications. Carriers now ask direct questions about AI deployment, but those answers can become stale because a company’s AI footprint changes faster than its policy period. A target that answered narrowly in the spring and deployed three new tools by the fall may have created an exposure that sits outside the policy. Nothing in a standard diligence request reaches the completed applications, so the exposure transfers without ever being priced.

Tail and runoff. Coverage purchased at closing carries the endorsements of the underlying program. The coverage purchased at closing carries the exclusions in the existing insurance program. Buying six years of continued D&O coverage does not restore coverage for excluded AI-related claims. It simply preserves the exclusion for six more years. The indemnity structure then absorbs risk both parties assumed insurance would cover, and it does so when the seller’s ability to fund it is most limited.

There’s a fifth point that cuts the buyer’s way and is worth raising in negotiation. Courts sometimes treat a new exclusion as evidence that earlier policy language was broader. In the AI context, for example, suppose a technology company had a 2024 liability policy with no AI exclusion, and its insurer added one when the policy was renewed in 2026. If the company were sued over an AI-related event that occurred in 2025, the buyer could argue that the 2024 policy should provide coverage because the conduct occurred while that policy was in effect and it contained no comparable exclusion. The later exclusion could also support the argument that the earlier policy covered risks the insurer later decided to exclude.

That possibility makes the target’s older insurance policies worth reviewing before closing, not just the policies in effect on the closing date. An older policy may cover a claim arising from something the target did years earlier, even though the same conduct would be excluded under its current policy. The buyer should therefore consider preserving those older policies and the records showing their terms rather than allowing them to be canceled or discarded at closing.

Representation and Warranty

Representation and warranty insurance, which protects a buyer against certain losses resulting from a seller’s breach of its representations and warranties, does not necessarily solve an AI insurance gap. Instead, that gap becomes a pricing problem rather than a drafting problem.

RWI underwriters are scrutinizing AI-specific representations closely and, where the diligence record is thin, pushing for broad exclusions reaching the development, deployment, governance or performance of the target’s AI systems generally. Skadden has made a similar observation about data provenance and model performance representations.

Combined with the underlying coverage problem, the consequence can be significant. An acquiring company that accepted deal risk in reliance on the RWI policy may discover that the target’s insurance no longer covers the AI-related loss because of an ISO exclusion. It may then also find that the RWI policy does not cover the loss, depending on its terms. What looked like an insured risk can therefore become a risk the buyer must bear itself. The exclusion isn’t necessarily reflected in the premium. More importantly, it can remove the insurance protection that helped make the risk acceptable.

The corollary is that AI governance documentation has deal considerations. A target that can produce a model inventory, vendor diligence records, testing documentation, human-review protocols, and incident-response procedures gives its insurers and the RWI underwriter evidence that it understands and manages its AI risks. A target that cannot provide that documentation may have difficulty obtaining coverage for the risks that matter most to the buyer. If the buyer cannot insure those risks, it may demand a lower acquisition price to compensate for the additional risk it is assuming, a larger escrow or a special indemnity that the seller must fund.

That convergence is worth explaining to clients on the sell side well before a process begins. The file a company assembles to satisfy its carrier at renewal is also the file a buyer will demand in diligence and an RWI underwriter will price. Regulatory requirements may change, but the underlying governance work doesn’t disappear. It migrates to the underwriting file.

Interim Period Problems

There’s a timing trap that no standard form addresses. Renewal dates don’t pause for a transaction.

A company that signs in October and closes in February may renew its coverage in the interim. The typical operating agreement, which restricts the target from materially altering its insurance or allowing policies to lapse, may not prevent it from accepting a renewal that carries an AI exclusion the prior policy didn’t have.

From the target’s perspective, nothing improper happened. It maintained continuous coverage on the terms its carrier offered. For the buyer, however, the diligence performed at signing has been overtaken, and the closing condition tied to the accuracy of the insurance representation may or may not catch it.

There are two straightforward fixes. Add renewal or replacement of any insurance policy to the list of interim actions requiring the buyer’s consent or, at minimum, require prompt notice with copies of the renewal terms and endorsements. And draft the insurance representation so the deal’s closing terms measure coverage scope rather than status, which means the representation has to reach exclusions in the first place.

The same logic applies to the applications. A target completing a renewal application during the interim period is making representations to a carrier about its AI deployment while simultaneously making representations to a buyer. The two sets should be consistent. Requiring copies of interim applications is a cheap way to find out whether they are.

What to Change

The fix is straightforward: make AI insurance coverage and AI risk part of the acquisition process from the start, rather than discovering gaps after signing or closing.

Expand the request list. Ask for the complete policy, all endorsements and the endorsement schedule for every line, not just general liability, plus completed applications and any AI-related questionnaires or supplements submitted to carriers. Ask for the same materials for the three prior policy years. Certificates and loss runs aren’t sufficient.

Redraft the representation. The insurance representation should warrant that none of the target’s policies contain an exclusion or sublimit relating to AI, machine learning or automated decision-making and that all information provided to insurers when the policies were obtained or renewed was accurate and complete. The second clause is important because inaccurate or incomplete information provided to an insurer could give the insurer grounds to challenge coverage. It is also likely to be one of the provisions sellers will negotiate hardest.

Add an AI representation that stands on its own. Cover the model inventory, training data provenance, third-party licensing terms, human oversight of consequential decisions and known incidents of AI error or hallucination affecting customers or employees. This is the representation the RWI underwriter will read most closely and its supporting diligence record determines whether the policy covers it.

Structure for the gap. Where the target materially deploys AI and its insurance excludes AI-related liability, address the risk separately in the purchase agreement rather than treating it as part of the deal’s general protections. The seller’s obligation to cover those losses should last long enough to address claims that could arise from the target’s pre-closing activities. Consider also whether the amount held back from the purchase price should be sized to the uninsured risk rather than to the customary percentage.

Raise RWI early. Underwriter appetite depends on the diligence record, which must exist before the underwriting call rather than be assembled in response to it. A seller that begins building the governance file at signing has already lost the pricing benefit.

Companies spent two years waiting to learn what AI compliance would require of them. The answer did not arrive from a legislature. It arrived from their broker, and it will come next in a buyer’s diligence request list.