Skip to content

The Record · Part 5 of 5

Down the Chain

Whose insurance pays when a sub leaks data to AI?

The Record, Part 5 of 5: Down the Chain. Whose insurance pays when a sub leaks data to AI?

Construction lawyers at Snell & Wilmer describe a subcontractor on a data center project for an AI company. The sub runs meeting summaries, takeoffs and supplier agreements through a rival AI company’s platform.

That alone, they write, could breach the prime contract’s confidentiality terms.

No one meant harm. The sub was trying to work faster. But you, the GC, may end up holding the risk.

The upload itself can be the breach

Prime contracts often define confidential information broadly: plans, specifications, pricing, schedules, even the contract terms.

Those confidentiality duties usually flow down to subs. In other words, your subcontracts pass the same promises along.

Snell & Wilmer’s Laurie Choi and Christian Fernandez warn that putting project documents into third-party AI tools can create “potentially substantial and uncapped liability.” Uncapped means no ceiling on what could be owed.

The usual damages limit may not apply

The waiver of consequential damages is the clause where the owner gives up claims for indirect losses, such as lost profits.

Many prime contracts make an exception to that waiver for confidentiality breaches. So an owner’s losses from a leak can be recoverable, and they can grow fast.

The sub’s promise to cover you can break

If the owner comes after you, you’ll turn to the sub’s indemnity. That’s its promise in the subcontract to cover losses it causes.

But what if the sub lacks the resources? Or its insurance excludes intentional data disclosures? An upload is a deliberate act, even when the leak isn’t. Then, the authors warn, the GC “could be left holding the bag.”

Insurance may not fill the gap, either. The authors add that standard commercial general liability (CGL) policies “may not cover this type of risk,” and professional liability (E&O) policies “may not be triggered either.”

Some GCs already require approval for AI

Some general contractors now add subcontract terms that bar AI on project work without prior written consent.

Under those terms, subs must name the tools they plan to use. They must also show the tools “do not retain or train on project data.”

The standard forms are behind

Jane Kutepova of Michelman & Robinson notes that AIA and ConsensusDocs forms “were not drafted with AI in mind.”

She recommends that contracts assign responsibility to “the party selecting or configuring the AI.” In short, whoever picks or sets up the tool should carry that responsibility.

What an AI rider should cover

Your carrier may ask about this, too. Lockton reports underwriters asking how AI is addressed in contracts with vendors.

A rider is a short add-on to your standard subcontract. Here’s what an AI rider should cover:

  • Restricted data. No bid pricing, confidential owner material, privileged material or personal information in AI tools without consent.
  • Disclosure. A schedule lists each approved tool, its account type and its training setting. The sub gives written notice before adding another.
  • No training. No training on your data, limits on how long it’s kept, and certified deletion when the work ends.
  • Human review. A qualified person reviews each AI-assisted deliverable and records the review. Using AI doesn’t lower the standard of care.
  • Insurance. The sub confirms that its CGL, professional and cyber policies don’t exclude AI-related claims, or discloses any that do. It also gives notice of new exclusions.
  • Everything else. Notice of incidents, flow-down to lower-tier subs, and verification on request.

A 90-day plan

Days 1 to 30: find it.

  • Name an executive owner for AI, and adopt a written AI use policy.
  • Survey project teams for every AI tool in use, including at outside firms.
  • Pull every insurance policy and endorsement, and flag any AI language.

Days 31 to 60: govern it.

  • Move work that commits the company to approved business-grade tools, or pause AI there. This includes prices, notices, claims and safety decisions.
  • Add the AI rider to new subcontracts and consultant agreements.
  • Test one high-stakes workflow on work you already know the answer to.

Days 61 to 90: prove it.

  • Start an error log by workflow. Track errors the AI introduced and errors your team caught.
  • Run a tabletop exercise on an AI data leak. That’s a sit-down drill where your team talks through how it would respond.
  • Before renewal, take your AI use policy, risk register and tool list to your broker. The risk register is a list of your AI uses and their risks, each with an owner.
  • Ask your broker for written answers on the shadow-AI question from Part 2. Is an employee’s or sub’s upload of confidential data to an AI tool covered?

What it means for a contractor

Your confidentiality promise to the owner is only as strong as your least-governed sub.

Ask your broker

Put this in writing: “If a sub’s AI upload breaches our owner’s confidentiality clause, which of our policies responds? Does our subcontract insurance requirement close the gap?”

The series in one line

AI tools will say yes to almost anything. Governance makes sure a qualified person says yes, too.

Thanks for reading all five parts.

Why we built SymphonIQ this way

Full disclosure: I run SymphonIQ, and this series is the argument behind it. Each part maps to a design choice:

  • The model will say yes. Ours is built to say “insufficient evidence” when the record is thin. Every number comes from a calculation tool, not the model.
  • Shadow AI. One private environment per customer. Nothing pooled, nothing trained across customers.
  • Who signed it? An authorized person makes the call, and the record shows who.
  • What underwriters ask. Every answer carries its receipt: the sources, the tools and the person who signed.
  • Down the chain. Every project is a sealed record, and nothing is kept if you leave.

If you want to see that on real work, bring us one live project.

Free resources: Our Model AI Rider for Subcontracts (Word or PDF) and AI Use Policy for Contractors are the two documents to start with.

Get The Record by email

Each part in your inbox the morning it publishes.

Then the occasional new guide or template for contractors. Unsubscribe in one click.

We confirm your address by email first. Privacy Policy

General information, not legal or insurance advice. Talk to your broker and counsel.

Sources

  1. Snell & Wilmer: Your AI Tool Could Be Your Next Breach of Contract, Aug 25, 2026
  2. Michelman & Robinson: Who’s Liable When AI Builds?, Nov 12, 2025
  3. Lockton Spotlight: AI, July 2026
  4. SymphonIQ: AI risk framework for contractors (90-day plan)

About The Record

The Record is written by Jared Smith, founder and CEO of SymphonIQ, which builds governed, single-tenant AI for engineering and construction.