Agent Authority
Home/Guides/Contracting

When an Agent Makes a Mistake

Contracting · Guide 5Last reviewed: November 2026About 12 minutes

Summary

  • A company is usually stuck with its own agent’s error. Undoing a contract for a one-sided mistake is hard, and in an early U.S. case, a court refused to let a business that knew the facts disown what its computer did.
  • UETA gives natural persons, as opposed to corporations and other entities, a way out of their own errors in automated transactions when the other side’s system gave them no chance to catch the mistake, if they promptly give notice, return what they received and haven’t benefited. That protection can’t be waived. A separate rule on agreed error-detection procedures is open to businesses too.
  • If an agent exploits an obvious error by the other side, such as a mispriced item, the other side may be able to undo the deal.
  • The best protection is prevention: confirmation steps, sanity checks and fast cancellation rights.

Five kinds of agent error

“The agent made a mistake” covers different problems, and the law treats them differently.

ErrorExampleMain rules
Bad inputA buyer tells its own agent to order 1,000 units instead of 100, and the agent sends that order accuratelyThe contract and ordinary mistake rules govern this input error. UETA’s individual error rule covers certain mistakes a person makes when dealing with the other party’s electronic agent.
Misread instructionsThe agent reads “per month” as “per year”Mistake; the tool view makes the company bear it
Bad judgmentThe agent picks a poor deal within its limitsUsually no remedy; it’s a bad bargain, not a mistake
MalfunctionA software bug repeats an order 40 timesMistake; contract terms, such as agreed error checks; UETA’s security procedure rule only if the error arose in transmission
The other side’s errorThe agent buys laptops priced at $6 instead of $600Unilateral mistake by the seller; knowledge of the buyer

UETA’s error rules

UETA § 10 has two error rules and a fallback.

  • Agreed security procedures. If a change or error occurs in transmission between the parties, the parties agreed to use a security procedure to detect changes or errors, and one side didn’t use it when doing so would have caught the error, the side that followed the procedure may avoid the effect of the error.1 The rule covers errors that arise in transmission. An agent’s wrong decision, accurately sent, isn’t one.
  • Individuals in automated transactions. An individual may avoid certain errors they make in dealing with the other side’s electronic agent. For example, a sole proprietor enters the wrong quantity in a supplier’s ordering system and has no opportunity to prevent or correct the error. On discovering it, the buyer must promptly tell the supplier about the error and that they did not intend to be bound, and take reasonable steps to return what they received or destroy it if instructed. The buyer must also have received no benefit from it and made no use of it.2
  • Everything else is governed by other law, including the law of mistake, and the parties’ contract.3

The individuals rule and the fallback to other law can’t be changed by agreement.4

Mistake under contract law

Outside UETA § 10’s two error rules, the general law of mistake applies. It’s demanding.

  • One-sided mistake. A party can undo a contract because of its own mistake only if all of these are true: the mistake was about a basic assumption of the contract; it materially affects the exchange, to that party’s disadvantage; the party doesn’t bear the risk of the mistake; and either enforcing the contract would be unconscionable, or the other side had reason to know of the mistake or was at fault in causing it.5

    A party bears the risk if, when it made the contract, it knew its information about the relevant facts was limited and treated that information as sufficient.6 For an automated purchase, the question is what the company knew when the order was placed and whether it chose to proceed on that basis.

  • Machine errors. In Bockhorst, an insurer couldn’t avoid a reinstatement its computer issued in error, because its employees already knew the facts. The error was the company’s, not the machine’s.7

For a company, the practical result is that its agent’s errors are usually its own. If the company can’t meet the mistake test, it can still use any cancellation or return right in its contract or ask the counterparty to unwind the deal.

When the other side made the error

Agents are good at spotting bargains, including ones that result from the other side’s mistakes. Two cases from Singapore show how courts approach this.

In Chwee Kin Keong v. Digilandmall.com, a website listed laser printers at a tiny fraction of their price. Buyers who ordered in bulk, knowing the price was wrong, couldn’t enforce the contracts.8 In Quoine v. B2C2, trades executed by algorithms at abnormal prices were upheld. For contracts made by fixed algorithms, the court said the relevant knowledge may be the programmer’s or that of the person running the algorithm, judged from when the program was written until the contract was formed, and found the programmer hadn’t known of or exploited the error.9

Together they suggest a rule of thumb. Singapore law requires actual knowledge for unilateral mistake at common law (or constructive knowledge plus unconscionable conduct in equity), whereas U.S. contract law under Restatement § 153 allows avoidance where the other side merely had reason to know of the mistake. If your agent takes advantage of an obvious error, the deal may not stand, and designing the agent to hunt for such errors makes it less likely to. Quoine dealt with systems that follow fixed rules. How to apply it to an agent that reasons its way to a decision is not yet settled.

Worked example

Gamma Ltd’s agent restocks office supplies. A catalog update changes a product’s unit from “each” to “case of 50,” and the agent, instructed to buy 200 units, orders 200 cases: 10,000 items. The supplier ships.

  • UETA § 10? No. The security procedure rule doesn’t apply because Gamma and the supplier didn’t agree on one, and because the error was the agent’s own choice, accurately sent, not an error in transmission. The individuals rule doesn’t apply because Gamma is a company, and the error was made by its own agent, not by Gamma in dealing with the supplier’s agent. So the general law of mistake decides.
  • Unilateral mistake? Weak. Only Gamma was mistaken, so it must meet the one-sided mistake test. If Gamma knew it had incomplete information about the catalog units and treated that information as sufficient when it ordered, it bears the risk. The fact that its agent ordered without review does not establish what Gamma knew. Enforcing an order for office supplies it can return or resell is unlikely to be unconscionable. Its best arguments are that the supplier had reason to know of the mistake, if the order was far outside Gamma’s usual quantities, or was at fault, if it changed the catalog unit without clearly flagging it.
  • Practical outcome: Gamma relies on the supplier’s return policy or negotiates. With a cancellation clause and a quantity cap enforced at the ordering system, the order would not have gone through.
For the engineering team: before an order goes to the supplier, have the ordering system compare it with past orders and hold any order that exceeds a set multiple of the usual quantity or value until a person approves it.

Preventing and handling errors

  1. Sanity checks: block orders far outside normal quantity, price or frequency.
  2. Agreed error checks: with key counterparties, agree on checks for unusual quantity, price and frequency, and on who may avoid an order the checks would have caught. That right works as a contract term. UETA’s first rule adds a statutory remedy only for changes or errors in transmission.
  3. Cancellation rights: negotiate a window to cancel agent-placed orders. See Agent Use Clauses.
  4. Act fast: when an error surfaces, notify the other side at once, and don’t use what was delivered.
  5. Don’t build agents to exploit errors. It increases the chance the deal will be undone and invites disputes.

Common situations

In each of these three situations, an agent’s error leaves a company asking whether it is bound and whether it can undo the result.

The company’s agent told a business customer something wrong. Is the company liable?

Short answer: The company can be liable for it, though its contract with the customer may limit what the customer can recover. The clearest decision so far is Moffatt v. Air Canada (2024), in which a Canadian small-claims tribunal held an airline responsible for wrong refund information from its website chatbot, and rejected the argument that the chatbot was responsible for its own statements.10 The customer there was a consumer, and the decision isn’t binding precedent, even in British Columbia. It doesn’t bind U.S. courts either, though it may be cited as persuasive.

The tribunal treated the chatbot as part of the airline’s website, not a separate entity, and that reasoning doesn’t depend on who the customer was. Between businesses, the contract usually does more of the work. An entire agreement clause, a disclaimer of statements outside the written terms, or a limitation of liability can narrow what a business customer recovers for a wrong answer, though how far those clauses reach a misrepresentation claim depends on their wording and the governing law. Without them, a business customer that reasonably relied on the agent’s answer is in much the same position as the customer in Moffatt.

  • If you deploy the agent: Review what the agent is able to promise, especially on price, credits, delivery and contract terms, and keep its answers tied to your written policies and the customer’s contract. Check what your customer contracts say about statements outside the written terms. Don’t assume correct information elsewhere will cancel out a wrong answer.
  • If you provide the agent: Customers will ask you to stand behind what it says. Limit that to failures of documented controls, and offer ways to tie answers to the customer’s approved content.

The agent issued a refund or credit, or changed an account. Is the company bound?

Short answer: If the company’s agent issued the refund or credit, check what the customer was owed and whether the company can recover a mistaken payment or reverse a credit. For an account change, verify who requested it as well as which system carried it out.

Refunds and credits. A refund or credit the agent issues is value the company gives up. The customer contract matters for both. If the company paid a cash refund by mistake, it can generally seek restitution to the extent the payment was not due.11 For a noncash account credit, check the contract for a right to reverse the entry.

Promises. If the agent promises a refund or credit it can’t issue, the company may still be liable for it: in Moffatt, the airline was liable for negligent misrepresentation and had to pay the difference between the fare paid and the bereavement fare. Whether a promise is also enforceable as a contract term is a separate question.

Tool or agent. Both views usually lead to the same place. Under the tool view, the refund log can show that the company’s agent issued this refund. Under agency rules, directing customers to that service channel can give them reason to believe the agent handles refunds.

Account changes. When a request to change a business customer’s contacts, billing details or payment details arrives, often from the customer’s own agent, the question is whether it really came from the customer. The company’s verification procedure is the main evidence of that. A weak one makes it harder for the company to show the request was the customer’s, though it doesn’t by itself decide who bears a loss. Requests to change payment details are a common route for fraud between businesses, so treat them as high risk whether a person or an agent sent them.

The records that decide disputes over refunds, credits, and account changes are the refund and credit policy and the agent’s limits; the conversation, including anything the agent promised; how the requester was verified; and the logs of the change.

  • If you deploy the agent: Cap refunds and credits per transaction and per day, require a person’s approval above a threshold, verify the requester before account changes, guard against instructions hidden in customer messages, and alert on unusual patterns.
  • If you provide the agent: Customers will ask what refund, credit and account change limits your agent enforces, and whether instructions hidden in customer messages can push it past them. Document the controls, and say who bears the loss when they fail.

The company’s travel agent booked a nonrefundable stay. Is the company bound?

Short answer: Usually, if the company’s travel agent made the reservation on terms the hotel accepted. The card and booking logs help establish who made it. A policy cap the company never communicated to the hotel usually offers little protection. Whether the booking matched what the traveler asked for matters more inside the company than against the hotel.

An employee asks the company’s travel agent to “find me a hotel under $400 near the client’s office.” The agent books a nonrefundable four-night stay at $450 a night on the company card, above the travel policy’s cap.

Against the hotel: under the tool view, the booking can be attributed to the company if it deployed the travel agent and that agent made this reservation. The company’s card record is evidence of who booked the stay. Its effect depends on the booking terms and applicable law. Under agency rules, the company is bound if the booking was within the agent’s authority, or if the hotel could reasonably believe it was, based on what the company did, such as letting the agent book with its card.12 A nightly cap in a travel policy the hotel never saw usually won’t help.

Against the card issuer: for a corporate card, the card agreement and network rules usually decide disputed charges, because Regulation Z’s billing error rules don’t apply to business purpose cards.13 Its limits on liability for unauthorized use generally do apply, though an issuer that issues ten or more cards for an organization’s employees may agree with the organization on different liability, so check the card agreement. Even then, employees are protected against liability for unauthorized use by someone else.14 But a charge the company’s own agent made, for a stay its employee used, is hard to call unauthorized. Regulation Z defines unauthorized use as use by someone without actual, implied, or apparent authority from which the cardholder gets no benefit.15

Inside the company: $450 a night plainly broke the employee’s “under $400” instruction; whether four nonrefundable nights fit what the employee asked for, and what the travel policy allowed, are less clear-cut. The answers, together with who configured the agent and the terms of the company’s contract with the agent’s vendor, decide who bears the cost. They don’t decide whether the hotel gets paid.

  • If you deploy the agent: Put the travel policy’s caps and refund rules into the booking system, not only in the agent’s instructions, and require a person’s approval for nonrefundable bookings or anything above a cap. Keep a word-for-word record of the traveler’s request.
  • If you’re the hotel or booking site: Show refund terms in a form agents can read, and keep the terms shown and the confirmation received for each agent booking; they’re your evidence in a dispute. A corporate rate agreement is the natural place for a business customer to tell you its limits.
  • If you provide the agent: Have the agent confirm with the traveler before booking anything nonrefundable or above the policy cap, and log the traveler’s instruction word for word.

Both sides of the deal

If you deploy agents

  • Add sanity checks on quantity, price and frequency.
  • Negotiate cancellation windows for agent-placed orders.
  • Act fast: notify the other side and don’t use what was delivered.

If you build agents

  • Build confirmation steps and anomaly thresholds as product features.
  • Say in your contract which errors are yours (controls that failed to work as documented) and which are the customer’s (its instructions and settings).
  • Don’t design agents to exploit other parties’ pricing errors.

Open questions

  • Whose knowledge counts when an agent that reasons for itself spots a pricing error: the programmer’s, the deployer’s, or the agent’s?
  • Does deploying an autonomous agent mean a company has assumed the risk of its mistakes?
  • If an agent with authority to issue refunds or credits gives one outside policy, can the company recover it as a payment made by mistake, or is it bound by its agent’s decision?
Sources
  1. Unif. Elec. Transactions Act § 10(1) (opens in a new tab) (Unif. L. Comm’n 1999). ↩
  2. Id. § 10(2) & cmt. 4; id. § 2(12) (defining “person” to include “an individual” and, separately, a “corporation”). ↩
  3. Id. § 10(3). ↩
  4. Id. § 10(4). ↩
  5. Restatement (Second) of Contracts § 153 (opens in a new tab) (Am. L. Inst. 1981). ↩
  6. Id. § 154(b) & cmt. c. ↩
  7. State Farm Mut. Auto. Ins. Co. v. Bockhorst (opens in a new tab), 453 F.2d 533 (10th Cir. 1972). ↩
  8. Chwee Kin Keong v. Digilandmall.com Pte Ltd (opens in a new tab), [2005] SGCA 2 (Sing.). ↩
  9. Quoine Pte Ltd v. B2C2 Ltd (opens in a new tab), [2020] SGCA(I) 2 (Sing.). ↩
  10. Moffatt v. Air Canada (opens in a new tab), 2024 BCCRT 149 (B.C. Civ. Res. Trib.). ↩
  11. Restatement (Third) of Restitution & Unjust Enrichment § 6 (opens in a new tab) (Am. L. Inst. 2011). ↩
  12. Restatement (Third) of Agency §§ 1.04 cmt. e, 2.03 (opens in a new tab) (Am. L. Inst. 2006). Unif. Elec. Transactions Act § 9(a)–(b) (opens in a new tab). ↩
  13. Regulation Z, 12 C.F.R. pt. 1026, Supp. I, cmt. 3(a)-2.i (opens in a new tab). ↩
  14. 12 C.F.R. § 1026.12(b)(1)(ii) (opens in a new tab); Supp. I, cmt. 3(a)-2.i (opens in a new tab). If an issuer provides 10 or more cards for an organization’s employees, it and the organization may agree to different terms, but not to hold an employee liable beyond these limits for someone else’s unauthorized use. § 1026.12(b)(5) (opens in a new tab); Supp. I, cmt. 12(b)(5)-2 (opens in a new tab). ↩
  15. 12 C.F.R. § 1026.12(b)(1)(i) (opens in a new tab); Supp. I, cmts. 12(b)(1)(ii)-1, -3 (opens in a new tab). ↩