Force Majeure, Indemnification, and What AI Actually Flags in a Review

Contract review is a search for exceptions: clauses that deviate from your standard, that shift risk in ways you didn't intend, that will matter if the deal goes sideways. Two clause families produce most of the work — force majeure and indemnification. They are also the two where AI flagging is at its most useful and, simultaneously, its most incomplete. This article walks through what each clause actually does, the red flags that matter, and the honest boundary of what a machine can find.

Force majeure: the clause that got famous in 2020

Before the pandemic, most lawyers skimmed force majeure. Then every litigator in the country learned the same lesson at once: force majeure is a contract-based defense, construed narrowly, and it excuses performance only for events the clause actually lists. It is not a general impossibility or frustration catch-all. If a pandemic is not in the clause, a pandemic is not covered.

Anatomy of a force majeure clause

• The event definition. The heart of the clause: which events count — commonly "acts of God, war, terrorism, strikes, governmental orders," and increasingly, post-2020, "pandemic or epidemic."
• The causation standard. Does the event need to prevent performance, or merely delay or hinder it? "Hindered" is a much weaker trigger and favors the party seeking relief.
• Notice. A deadline to notify the other side (commonly a set number of days), with consequences for missing it.
• Mitigation. A duty to use reasonable efforts to resume performance.
• Duration and termination. If the event lasts beyond a defined period (often 60–90 days), either party's right to terminate.
• Exclusions. The most important part. Most well-drafted clauses exclude the payment of money from what can be excused — because a party can almost always pay, even in a crisis.

The COVID record, in two cases

The pandemic produced a body of decisions that reads like a drafting seminar. In 1600 Walnut Corp. v. Cole Haan Co., 530 F. Supp. 3d 555 (E.D. Pa. 2021), a tenant argued that government shutdown orders excused its rent. The court disagreed: the lease's force majeure clause suspended performance for listed events, but expressly did not relieve the tenant of its obligation to pay rent — so the rent was still owed. In In re Hitz Restaurant Group, 616 B.R. 374 (Bankr. N.D. Ill. 2020), by contrast, a bankruptcy court held that a government stay-at-home order triggered a lease clause covering "laws, governmental action or inaction, orders of government," and partially excused the rent. The Hitz result is widely treated as the outlier; the rule the cases together teach is plain: if you want a pandemic trigger, write it in; if you are the party receiving payment, keep money out of the clause.

Force majeure red flags

• An event list that omits the risks that actually threaten your business — no "pandemic or epidemic," no "governmental orders."
• A causation standard so loose ("hindered") that almost any disruption qualifies.
• No notice deadline, no mitigation duty, no duration cap before termination.
• Payment obligations not excluded — or, conversely, a clause that lets force majeure excuse money obligations.
• For sellers of goods, remember UCC § 2-615 governs commercial impracticability separately — the contract clause and the statute interact.

Indemnification: the clause lawyers fight over most

Indemnity is risk allocation: one party agrees to carry defined losses so the other doesn't have to. It sounds simple; it is where deals routinely hang. The structure has four parts — who indemnifies whom, for what trigger, covering which losses, subject to what carve-outs.

Third-party claims vs. first-party claims

The classic indemnity covers third-party claims: a stranger sues your client because of the other party's product or conduct, and the other party pays the judgment and fees. The trap is the first-party indemnity — language like "Party A shall indemnify Party B for all losses arising out of this Agreement" — which quietly converts an ordinary breach remedy into an indemnity. That matters because indemnities often carry special rules and, critically, often carve out of the liability cap (more on that below). Courts generally presume indemnity covers third-party claims and require clear language to extend it to direct claims.

Duty to defend vs. right to defend

Two distinct things are often bundled in one sentence. The duty to defend obliges the indemnitor to step in and fund the defense immediately, triggered by the allegations, before liability is proved — broader than the duty to pay. The right to defend is the indemnitor's privilege to control counsel, strategy, and settlement. A red flag is either half missing: an indemnity with no defense machinery, or a right to settle that doesn't require the indemnitee's consent where an injunction or admission of fault is involved.

IP and data indemnities

In technology and SaaS contracts, the two most contested indemnities are intellectual property infringement (the vendor indemnifies the customer if the software infringes a third party's rights, with a right to procure replacement or terminate) and data security (indemnity for breaches arising from the vendor's platform). A missing IP indemnity in a software deal is a classic AI-detectable gap — and a classic negotiation battleground.

Enforceability limits

Not everything is indemnifiable. Many states — particularly in construction and oil and gas — have anti-indemnity statutes that void clauses requiring a contractor to indemnify an owner for the owner's own negligence: the Texas Oilfield Anti-Indemnity Act, Tex. Civ. Prac. & Rem. Code § 127.003, and the Texas Construction Anti-Indemnity Act, Tex. Ins. Code § 151.102, are the best-known examples. And across virtually every jurisdiction, you cannot contract away liability for your own gross negligence, recklessness, or intentional misconduct. These limits are jurisdiction-specific — which is exactly why a clause that looks benign in one state can be unenforceable in another.

Indemnification red flags

• Unilateral indemnity. Your client indemnifies everything; the other side indemnifies nothing reciprocal.
• Broad causation. "Any losses arising out of or relating in any way to this Agreement" — no link to fault at all.
• Missing IP or data indemnity in a deal where the risk plainly exists.
• No notice, tender of defense, or cooperation obligations — machinery without which the indemnity is unworkable.
• The cap interplay ignored — see below.

The indemnity–cap interplay: the flag AI can find but only you can judge

Here is the most instructive example of where AI flagging stops and lawyering begins. Most commercial contracts cap total liability — commonly at one times annual fees in B2B deals. The market compromise is that general indemnities sit inside the cap, while specific high-risk indemnities are carved out of it — IP infringement, confidentiality and data security breaches, gross negligence, and bodily injury or property damage are the usual carve-outs. An indemnity that is subject to the same cap as everything else is, for the risks that matter, hollow. An indemnity that escapes the cap entirely is the reverse risk.

An AI review tool will reliably flag whether an indemnity exists and whether it references the cap. It will tell you the carve-out language is present or absent. What it cannot do — what no model can do — is tell you whether this deal justifies an uncapped IP indemnity, whether the counterparty's willingness to give one changes the commercial balance, or whether your client's insurance and the counterparty's creditworthiness make the exposure theoretical. Those are judgment calls about the transaction, and they belong to the lawyer.

What AI actually flags — and what it still needs you for

Within its lane, the machine is excellent. Encoded with your playbook, an AI scan will, across an entire portfolio:

• Locate and classify every force majeure and indemnity clause with consistent attention — no tired reading on page 37.
• Flag a force majeure clause with no pandemic or epidemic trigger, if that is your rule.
• Flag a unilateral indemnity, a missing IP indemnity in a software deal, or an indemnity with no notice or control-of-defense provisions.
• Rate each deviation for severity and explain it, against your own standards.

What it still needs you for is the part that litigators and deal lawyers know best: the interplay between clauses, the jurisdictional context that can void an indemnity, and the commercial judgment about whether a flagged deviation actually matters in this deal. The reliability research reviewed in our explainer of the review pipeline applies here without modification: grounding reduces hallucination but does not eliminate it, and courts have sanctioned lawyers for unverified AI output. A flag is a hypothesis; verification is the practice of law.

A playbook example

This is what the workflow looks like with real rules. A firm might encode these standards for a portfolio of supplier agreements:

• Force majeure must expressly include "pandemic or epidemic" and "governmental orders," must require notice within 14 days, and must not excuse the payment of money.
• Indemnities must be mutual for IP infringement and confidentiality breaches.
• Liability cap at one times annual fees, with carve-outs for IP indemnity, confidentiality and data security, and gross negligence.
• Every indemnity must include notice, tender of defense, and cooperation obligations.

Run sixty contracts against those rules and the scan returns an exception list, rated by severity — contract 12 has no pandemic trigger, contract 27's indemnity is unilateral, contracts 4 and 31 cap their IP indemnities. The lawyer then works the list in order of materiality, opening each flagged clause in the source document to confirm the flag and judge the risk. That is the division of labor at its most defensible: the machine guarantees nothing gets missed; the lawyer decides what matters.

How Lawyer Assistant implements this

This is the pattern Lawyer Assistant's compliance playbook scan is built around. You define the rules above in plain language, the app classifies every clause in the document and checks each against your playbook, and findings come back with severity ratings and explanations — each one linked to the source text so your verification step is a click away. Because everything runs locally on your machine, the confidentiality of the contracts themselves is preserved by architecture, not by promise.

The bottom line

Force majeure and indemnification are where contract risk actually lives, and they are precisely the clauses where the review workflow matters most. Use the machine for what it is: a tireless, consistent classifier and comparer that turns a stack of contracts into a rated exception list. Keep for yourself what it cannot touch: the interplay between clauses, the law of the jurisdiction, and the judgment about which exceptions are worth fighting for. The tools get better every year; that boundary does not move.

"The machine finds the exceptions. The lawyer decides which ones are the deal."

Sources & further reading

• 1600 Walnut Corp. v. Cole Haan Co., 530 F. Supp. 3d 555 (E.D. Pa. 2021).
• In re Hitz Restaurant Group, 616 B.R. 374 (Bankr. N.D. Ill. 2020) — widely cited as the outlier reading of a broad government-orders clause.
• UCC § 2-615 — commercial impracticability for contracts for the sale of goods (text).
• Texas Oilfield Anti-Indemnity Act, Tex. Civ. Prac. & Rem. Code § 127.003 (text); Texas Construction Anti-Indemnity Act, Tex. Ins. Code § 151.102 (text).
• ABA Formal Opinion 512, "Generative Artificial Intelligence Tools" (July 29, 2024) — competence, confidentiality, candor, and supervision when lawyers use AI (full text (PDF)).

This article is general information about technology and professional practice. It is not legal advice for any specific matter, and rules vary by jurisdiction — verify against the authority applicable to your matter.