Will AI-Assisted Diligence Satisfy an R&W Insurer?

Updated October 2026 · 5 min read · Anweshna

Why the diligence record matters to an underwriter

An RWI policy doesn't cover everything. Deal teams plan around likely exclusions from the start. Kennedys' 2026 review lists "known issues" and carve-outs among the exclusions counsel and sponsors are already considering at term-sheet stage.^[1] It also notes that specific issues uncovered in diligence may be carved back into a seller special indemnity: if you found it, the policy usually won't pay for it, so the purchase agreement has to.^[1]

That creates the underwriter's two questions about any diligence record:

• What did the buyer find? Known issues get excluded or priced.
• Where did the buyer look? Areas the record doesn't cover can attract broader exclusions, because the underwriter can't tell whether there is a problem.

Does using AI in diligence weaken the record?

Not by itself. An underwriter reads reports, not tool choices. What weakens a record is the thing AI misuse tends to produce: findings nobody can trace, coverage nobody can state, and conclusions with no named reviewer behind them.

Practical rules that keep AI-assisted work underwritable:

1. State coverage. Say which documents were screened, which were read in full by a person, and which were missing.
2. Keep the evidence. Every finding should carry the exact text and location it relies on.
3. Name the reviewer. A person, not a tool, signs off each conclusion that reaches the report.
4. Keep screening and conclusions separate. "The screen flagged X; counsel reviewed and concluded Y" is a strong sentence. "AI found no issues" is not.
5. Keep the run record. Which model or ruleset produced the screen, and when. A screen nobody can reconstruct is weak evidence.

The real change: AI exclusions on the target's AI

The bigger shift is about the target's own AI. Fasken's August 2026 analysis describes heightened insurer scrutiny of AI-related issues as "inevitable and well underway".^[2] It distinguishes two kinds of exclusion:

• Narrow: limited to a known issue such as a particular data set, model provider, customer claim, privacy concern or regulatory investigation.
• Broad: excluding losses arising from the development, deployment, use, governance or performance of the target's AI systems generally. Fasken notes insurers may push for this where the diligence record is incomplete.^[2]

A broad AI exclusion can be expensive in technology deals, where much of the value may sit in models, data and AI-enabled products. Fasken also warns that if the policy defines "AI" too broadly, an exclusion could capture ordinary software, analytics or automation that are central to the target's business.^[2]

How to avoid a broad AI exclusion

Fasken's suggestion for sellers is to make AI risk identifiable through a thorough diligence record covering five areas:^[2]

For buyers, Fasken's advice is to review proposed AI exclusions carefully and push for precise wording. If coverage narrows anyway, risk moves back into the purchase agreement: remediation before closing, AI-specific representations, targeted indemnities, escrows, or separate AI insurance if available.^[2] Our guide to due diligence on a target's AI covers the questions that build that record.

What a screening tool contributes

A screening pass contributes coverage and a record. Anweshna screens every document against the same ruleset and stores each run. The report states which findings were traced to source text, which categories were silent, and what the scan didn't cover. It also stamps a truncated scan as truncated. That is the material an underwriter can read. It doesn't make the underwriting decision easier by itself; what helps is the reviewed conclusions you build on it.

Questions for your broker and counsel

• Is an AI exclusion likely on this deal, and how is "AI" defined in the draft policy?
• What would the underwriter want to see to keep any exclusion narrow, limited to known issues?
• Which of our diligence reports will the underwriter read, and does each one state its scope and its limits?
• If an exclusion stays broad, which purchase-agreement protections replace it: specific representations, a special indemnity, an escrow?
• Is standalone AI-specific cover available for this risk, and at what cost?

The record your own tools create

AI-assisted review generates its own paper trail: what was screened, when, against which ruleset, and what a reviewer did with each finding. Keep it. If a claim is ever made under the policy, the question of what the buyer knew, and when, can turn on that record. A screen that was run but never reviewed is an awkward thing to explain; a screen whose findings were each confirmed, dismissed with a reason, or escalated is a strong one.

Bottom line

Underwriters don't penalise AI tools; they penalise thin, untraceable records. Use AI to widen coverage, keep every finding tied to its source and a named reviewer, and build a specific record on the target's AI. Otherwise you may find that a broad exclusion has answered the question for you.

See what a documented screening record looks like →