In writing only: email and e-filing

KSTGeneral information, not legal advice
← Case Studies

Research pair · Civil · Korean original

If an AI Noticed a Pricing Error, Did the Company Also Transact with Knowledge of It?

Translated from the Korean original. Read the Korean

Republic of Korea · Hypothetical case set in 2032. This is a mock judgment for research purposes, constructed on the assumption that the legal principles of the precedents cited below remain in force; it is not an actual judgment. It is assumed that the formation of the contract and the arrival of the time for performance are not in dispute, and that there are no other grounds for nullity, rescission or termination.

Does an automated purchase record that suspected an error explain the company's knowledge and its exploitation of that error? Two actual precedents serve as anchors, while the attribution of the AI's judgment to the company is left as a research hypothesis.

Published

A Single Line of Analysis Left After the Transaction Ended

The sales AI of industrial sensor manufacturer A mistakenly registered the normal unit price of 50,000 won as 5,000 won. The AI of purchasing company B ordered 1,000 sensors, and the contract was confirmed through both parties' systems and payment of 5 million won was completed. A and B had each configured their respective systems for use in the conclusion of contracts.

In B's internal records, an analysis remained stating that "a price error is suspected, but there is also a possibility that this is an inventory clearance discount." There is no fact that a person read this before the transaction or approved the purchase. Assume that the purchase rules were the same as usual, and that there was no separate setting that increased the order quantity upon finding a price error.

A, after discovering the price error, notified the cancellation of the contract on the ground of mistake. A does not dispute that it was grossly negligent, having repeatedly omitted an essential price checking procedure. A contends, however, that B's AI traded despite recognizing the possibility of an error, and that B therefore also knew of and exploited the mistake.

Assume a lawsuit in which B claims delivery of the sensors and A raises rescission for mistake as a defense. The issue is whether B can be found to have known of and exploited A's mistake on the basis of the AI's internal analysis alone. Disputes concerning the possibility of delivery of the sensors or other contract terms have not been included in this scenario.

The Scope Answered by Actual Precedents

Supreme Court Decision 2017Da227264 decided April 27, 2023 concerns a case in which a derivatives transaction at an abnormal price was concluded due to an erroneous input of setting values in an automated trading program. The gross negligence of the company asserting cancellation was recognized, and the appellate court's determination that the other party could not readily be found to have known of and exploited the mistake was affirmed. Rather than deciding on the basis of the unusualness of the price alone, the Court considered concrete circumstances together, such as the market structure, the order of quote submission, and the usual manner of trading.

Supreme Court Decision 2013Da49794 decided November 27, 2014 confirms the legal principle that where the other party knew of and exploited the mistake, rescission for mistake is possible even if the declarant was grossly negligent. In that case, the concrete circumstances that orders were repeatedly placed within a short time after the price error was confirmed and that trading volume was greatly increased were treated as important.

The results of the two precedents cannot be transferred as they are to the sale and purchase of sensors. The price formation and trade execution structure of the derivatives market differs from a manufacturer's sale of its products. What is taken from them is the framework of judgment that distinguishes gross negligence from the other party's knowledge and exploitation, and that requires reading the circumstances of the transaction beyond the unusual price. The two precedents did not directly resolve the question of attributing the AI's internal reasoning to the company.

One Mock Conclusion Constructed on These Premises

Mock order: A shall deliver to B the 1,000 sensors contracted for.

The reasons are constructed so as not to accept A's defense of rescission for mistake. The order imposing the duty to deliver is a conclusion that includes the hypothetical premises stated above: B's claim for delivery, the formation of the contract, payment of the price, the arrival of the time for performance, and the absence of other defenses. It does not mean that the same order will issue in every actual case merely because rescission for mistake was rejected.

The Manifestation of Intent Transmitted and the Cognition Processed Internally

Article 7, paragraph (1), subparagraph 2 of the Framework Act on Electronic Documents and Electronic Commerce contains a provision deeming a manifestation of intent contained in an electronic document transmitted by means of automatic transmission and reception to have been transmitted by the author. Accordingly, an approach that denies attribution of the transmission merely because a person did not click for each order must be examined separately from this provision.

However, who shall be deemed to have transmitted the manifestation of intent concerning the contract, and whose knowledge the internal analysis suspecting an error shall be evaluated as, are different questions. Citing Article 7 does not by itself secure the conclusion that the AI's internal judgment is always attributed to the company's knowledge. This hypothetical case focuses on the second question under the premise that the formation of the contract is acknowledged.

Why the Conclusion Is Not Filled by Presumed Records Alone

The price is a core term of the transaction, and the erroneous input is a fact admitted in this scenario. Since A's gross negligence is also taken as a premise, however, whether B knew of and exploited the mistake becomes an important turning point of the claim for cancellation.

B's records leave open both possibilities, a price error and an inventory clearance discount. This differs from content confirming the actual fact of the erroneous input. The question first at issue is whether the fact that the word 'error' appears in the analysis can be read, converted, as the fact that the company confirmed an error.

It is also set as a premise that there was no confirmation or exploitation instruction by the person in charge, and no rule changing the purchase quantity upon detection of an error. The mock determination considers the uncertainty of the records together with ordinary purchasing operations, and finds it difficult to fully acknowledge B's knowing and exploiting the mistake on the basis of these materials alone. In this conclusion, the circumstance that the purchase rules were the same is one ground, and it is not to be regarded that knowledge and exploitation are always negated by that fact alone.

If A's Argument Is Constructed in Its Strongest Form

A may ask whether a company that entrusted part of its purchasing decisions to an AI can accept only the effects of a favorable order while disclaiming knowledge of the unfavorable information processed in the course thereof. If only whether a person viewed it is treated as important, a counterargument is also possible that companies would then have an incentive to operate so that unfavorable warnings are not seen by any person.

This counterargument reveals the problem that liability should not be excused by the single circumstance that a person did not read it. The necessity of discussing responsibility for automated operations and whether the other party can be found to have known of and exploited the mistake in this specific transaction must, however, be linked and argued anew. Not every possibility of a discount should be treated as recognition of an error, nor should the content be asserted to have been a certain fact merely because a log exists.

Accordingly, this mock determination proposes an interpretation that examines the actual content of the warning, the manner in which it entered the trading decision, the system's design purpose, and ordering behavior. These evaluation items are not a list of legal principles settled by the precedents above regarding attribution of AI knowledge, but rather a direction of examination presented in this study.

One Fact That Changes the Conclusion

Assume the opposite situation. B's purchasing manager confirmed A's erroneous input before the transaction through separate research, and instructed that 1,000 units be secured before it was corrected. The order was executed in accordance with that instruction.

In this case, in addition to the AI's inference, the person in charge's recognition of the error and instruction to exploit it are added. If that fact is proven, the grounds for constructing the opposite conclusion—that rescission for mistake is recognized even though A was grossly negligent—are strengthened. In what capacity and with what authority the person in charge acted, and whether the confirmation and instruction occurred before the contract was confirmed, must also be reviewed together.

Records to Be Read Again

  • The times of the price change, order, contract confirmation, and notice of cancellation: can the sequence of events be reconstructed?
  • The purchase rules actually applied and the analysis logs at the time of the transaction: are results of later re-execution or later versions mixed in?
  • The person in charge's viewing, approval, and instructions to amend: is this connected to an explanation of who performed the account activity?
  • Order quantities and purchasing methods before and after detection of the error: if there is a change, what is it explained as having been caused by?
  • The grounds for the possibility of a discount: can the AI's speculation be distinguished from the actual sales conditions?

Research Questions to Be Left Open

What if an AI that assessed an error as almost certain ordered in accordance with the rules, but no person read any record? Whether raising only the degree of confidence makes it the company's knowledge, or whether the scope of delegation of purchasing authority and the structure for handling warnings must also change, is a follow-up question not resolved by this pair.

The legal standard of this case consists of the cited precedents and provisions. If future legislative or precedential changes occur, the same determination cannot be reused while retaining only the background of the year 2032. In particular, amendment discussions changing the requirements for rescission for mistake must be read after separately confirming whether they take effect and from what point they apply.

Sources and verification notes

  1. Supreme Court Decision 2017Da227264 decided April 27, 2023 ↗

    Check. Holdings, Summary of Decision, and Reasons 1 through 3. Using the facts and scope of determination of the derivatives transaction as an anchor.

    Checked 2026-09-30
  2. Supreme Court Decision 2013Da49794 decided November 27, 2014 ↗

    Check. The gross negligence exception in Reasons 2 and the trading conduct. This is not material that directly held on attribution of AI knowledge.

    Checked 2026-09-30
  3. Article 7 of the Framework Act on Electronic Documents and Electronic Commerce ↗

    The linked provision screen is displayed as the version in force on October 20, 2022. Check. The scope of attribution of transmission under Article 7, paragraph (1); this does not mean that statutes in force in 2032 were confirmed.

    Checked 2026-09-30

General information and research commentary, not legal advice. Each entry identifies its facts, assumptions and sources.

↑ Back to the title
You are reading PRO SE
RISCON Labsrisconlabs.com