Will R&W insurance exclude AI and training-data risk in your software deal?

What does R&W insurance cover in a software deal?#

Representation and warranty insurance covers a buyer's loss when a seller's representations in the purchase agreement turn out to be untrue, subject to the policy's retention, limit and exclusions. In software deals it often replaces much of the seller's indemnity, which is why sellers care about what it leaves out.

The underwriter does not repeat diligence from scratch. It reviews the buyer's diligence reports, asks follow-up questions on an underwriting call and then binds a policy. Matters the buyer already knew about, purchase price adjustments, covenants and forward-looking statements are common standard exclusions.

Deal-specific exclusions are different. They appear when the underwriter sees an area where diligence was thin or the risk is hard to size. AI and training data can fit that description in software deals, because the facts sit across contracts, code, data pipelines and privacy notices.

Why AI and training data attract exclusions#

AI and training-data risk attracts exclusions because one dataset can touch several representations at once and the evidence is scattered. The intellectual property, privacy, data security and material contracts reps can all be breached by data used without the right permission.

Typical concerns include models trained on customer data under terms that say nothing about training, datasets scraped from the web, third-party datasets with restrictive licenses, records licensed out to AI developers and AI vendor tools that may retain inputs. An underwriter facing these without documents has little choice but to exclude them.

Software sellers often sit on both sides. Many train product features on their own platform data, and some license operational records outward. Both activities belong in the diligence file.

Narrow vs broad AI exclusions#

A narrow AI exclusion removes coverage for a defined dataset, model or group of contracts, while a broad AI exclusion removes coverage for any loss connected to developing, training or using AI. The difference decides how much of the seller's AI exposure the buyer must cover some other way.

Sellers usually prefer the bottom rows of the table. Getting there depends less on argument than on what the diligence file can prove.

What underwriters and buyer counsel ask about AI and data#

Underwriters and buyer counsel ask about AI and data in a predictable order: what models exist, what trained them, what rights covered that data and what went out the door. Answers backed by documents move the conversation toward narrow exclusions.

• 1A register of models and AI features, with their purpose and where they run.
• 2Training data sources by category: customer product data, company records, licensed-in data, public or scraped data.
• 3The customer contract versions that apply to each source, including usage data and AI clauses.
• 4Any data licensed to third parties, with permitted use, term, exclusivity and privacy treatment.
• 5Privacy notices and processor commitments that touch training or analytics.
• 6Third-party AI tools in use and their terms on retention and training.
• 7Complaints, claims or customer objections about data use.

How a documented provenance record narrows exclusions#

A documented provenance record narrows exclusions because it lets the underwriter separate known, covered facts from the residual gap. Instead of excluding everything that touches AI, the underwriter can carve out only the dataset or cohort the record shows is uncertain.

Build the record per dataset, not per model. Each entry should be short enough to walk through on an underwriting call and specific enough to match a disclosure schedule.

The same record helps outside the policy. Buyers use it to scope any special indemnity, and it gives the seller a factual basis for resisting a broad exclusion.

Illustrative: a field service SaaS seller faces a broad exclusion#

Illustrative: a fictional field service scheduling software company is selling to a private equity buyer that plans to use R&W insurance. The company trained a route-suggestion model on customer job data and had separately licensed de-identified support tickets to an AI developer.

The first draft of the policy carried a broad AI exclusion. The company's counsel then produced a provenance record. It showed that customers on current terms had agreed to aggregated use of job data for service improvement, that the support tickets were company records released under a signed license with a privacy record, and that a group of legacy customers had signed older terms that said nothing about training.

The underwriter replaced the broad exclusion with a narrow one covering the route model's use of data from the legacy cohort. Buyer and seller handled that gap with a special indemnity, and the support-ticket license was reviewed like any other material contract.

Steps to take before the underwriting call#

Preparation before the underwriting call matters more than negotiation during it. Most of the work is assembling documents the company already has, in a form a stranger can follow.

• 1Ask deal counsel early whether the buyer intends to use R&W insurance and which reps cover data and AI.
• 2Build the model register and dataset provenance record before the data room opens.
• 3Map every customer contract version to the datasets it touches.
• 4Gather the documents for any outbound data license, including the privacy record and release approvals.
• 5Close gaps you can close now, for example by retraining a model without a disputed cohort.
• 6Agree with counsel how residual gaps will be disclosed rather than discovered.

Where SourceX records fit in diligence#

When a seller has licensed records through SourceX, the outbound side of AI diligence is already written down. Every package carries a SourceX Evidence Packet covering provenance, licensing rights, permitted use, the privacy record and release authorization, produced through the SourceX five-step transaction of Supply, Rights, Preparation, Approval and Delivery.

SourceX does not advise on insurance or deal terms, and the packet does not replace deal counsel or your broker. It gives them a consistent, dated record to work from.