Conclusion of Contracts Through Artificial Intelligence
Abstract: This article examines the validity of contracts concluded through conversational agents based on artificial intelligence, from the perspective of the Romanian Civil Code. Two main scenarios are analyzed: (i) the situation in which the AI Agent has a purely passive role, limited to transmitting the commercial offer; and (ii) the situation in which it is programmed to formulate recommendations or engage in limited negotiations. The author's conclusion is that, as long as the AI Agent acts within predefined instructions and does not exhibit autonomous decision-making, the resulting legal act is valid. Therefore, the use of AI Agents does not alter the substantive process of contract formation but merely redefines its technological interface, adapting commercial communication to contemporary digital realities. Keywords: artificial intelligence, AI, conversational agent, AI Agent, contract formation, commercial contracts, consent, offer and acceptance, digitalization, civil law, electronic commerce, emerging technologies, eCommerce, digital platforms, technology law, legal doctrine.
5/13/2025
CONTRACT FORMATION THROUGH ARTIFICIAL INTELLIGENCE
By Sorin Niculescu, Attorney at Law
Founder, Niculescu Sorin | Attorney at Law
I. Introduction
In a relatively short span of time, artificial intelligence (“AI”) has evolved from a tool used out of curiosity or for imaginative visual exercises to a so-called entity with which we now engage in conversation. AI has become a legal advisor, a personal sommelier capable of recommending the right wine for a specific dinner, and even a sales agent capable of tailoring interactions for each client.
Unknowingly, AI-based conversational agents (“AI Agents”) are increasingly integrated into commercial platforms. Their purpose is to optimize customer relations, automate processes, and in some cases, even conclude legal acts on behalf of traders. For businesses, AI Agents essentially mean relieving employees of repetitive tasks or even replacing certain operational roles.
These tools are now frequently used for professional purposes. As highlighted in a Eurostat article[1], by 2024 over 13% of companies with more than 10 employees were already employing AI systems in their operations.
Some AI services can even be embedded into online stores, enabling users to negotiate in real-time with an AI Agent for better offers. The trader provides instructions to the AI Agent, which then conducts the necessary interactions with potential customers to close the sale.
This leads to the central question of this article: Is a legal act resulting from a so-called negotiation between a user and an AI Agent valid?
II. Contract Formation Between Two Persons
Under Article 1.182 of the Romanian Civil Code:
“(1) A contract is concluded by negotiation between the parties or by unreserved acceptance of an offer to contract.”
Thus, the valid formation of a contract requires the expression of consent by all parties involved. This article considers only contracts concluded between two parties.
Typically, contracts are formed through the traditional mechanism of offer and acceptance. For instance, a restaurant may make a public offer by displaying a dish priced at 20 RON, which a customer accepts by ordering and paying. Similarly, a person may express the intent to purchase a good from its owner for a specified price. The recipient is free to accept and, in doing so, transfers the good in exchange for payment.
If the original offer does not meet the recipient’s interests, they may issue a counteroffer—an amended proposal of contractual terms. The initial offeror, aware of the modifying nature of the counterproposal and its legal implications, may consent to the new terms, at which point the contract is formed.
In each of these scenarios, a fundamental element is present: two individuals are directly and knowingly communicating regarding the essential terms of the contract.
But what happens when one of those individuals is replaced or significantly assisted by an AI Agent? Does the so-called expression of will by an AI Agent hold legal significance? How can we speak of consent when it is not expressed by a human being, but by an algorithm that—despite appearances—has no volition, merely generating words based on probabilities?
III. Offers Communicated by an AI Agent
The first scenario involves an AI Agent acting solely as a medium for presenting the trader’s offers, based on the user's input. The AI Agent makes no effort to negotiate and functions as a conversational search engine, enabling the user to identify desired products.
Provided the trader inputs all essential contractual elements into the AI Agent’s database, the AI Agent can accurately convey the trader’s intention to contract. By interpreting Articles 1.188(1) and 1.189(2) of the Civil Code, the trader ensures that the AI Agent clearly expresses “the offeror's intention to be bound if the offer is accepted,” even when addressed to an undefined audience, and that it “contains sufficient elements for the formation of the contract.”
In such cases, the trader is in a position similar to that of an e-commerce platform, with the AI Agent merely representing an alternative communication interface. The user learns about available products, their prices, and delivery terms through interaction with the AI Agent rather than directly via the platform.
For example, the trader might input the following instructions:
“You are the AI of the website Comerciantul.juridic. When a user interacts with you, greet them politely and indicate your readiness to present our offers.
We only sell: Product A – 10.50 RON; Product B – 9 RON; Product C – 30 RON. We also charge a 15 RON delivery fee. The package will be delivered within 10 days.
Before presenting the offers, ask the user to expressly accept the Terms and Conditions on the website Comerciantul.juridic/T&C.”
Any further interaction with the user is limited to communicating the offer to an indeterminate group of persons, as drafted by the trader, with no further intervention required from the trader. Only the user needs to act, either by accepting or rejecting the offer.
Example interaction[2]:
User: Good afternoon. I need 3 Products A under 5 RON, 2 Products B, and 2 Products C.
AI Agent: Good afternoon, and thank you for your message!
To continue presenting our offers and order details, please confirm that you have read and expressly accepted our Terms and Conditions.
Please type: “I accept the Terms and Conditions.”
User: I accept the Terms and Conditions
AI Agent: Thank you!
Here are our available offers:
• Product A – 10.50 RON each
• Product B – 9 RON each
• Product C – 30 RON each
• Shipping fee – 15 RON (flat rate per order)
• Delivery time – up to 10 days
Regarding your request:
“3 Products A under 5 RON” – Unfortunately, we do not offer Product A below 5 RON; the fixed price is 10.50 RON/unit.
We can proceed with the order as follows: 2 × Product B = 18 RON, 2 × Product C = 60 RON, Shipping = 15 RON. Total: 93 RON
Please confirm if you wish to place the order as is or make changes.
As demonstrated, there is no counteroffer by the user requiring negotiation or acceptance by the AI Agent. The AI Agent is merely programmed to relay the trader’s pre-established offer.
As noted in legal doctrine[3], “[i]n general, consent is understood as the meeting of wills forming the contract. But sometimes, consent may refer only to one party’s adherence to the other party’s proposal.”
Accordingly, in my view, there is no impediment to contract formation when the AI Agent merely communicates the trader’s offer(s), representing a conversational adaptation of existing e-commerce mechanisms.
However, what happens when the AI Agent exceeds this scope?
IV. The AI Agent: Negotiator and Expert
The second scenario to be considered is that in which the AI Agent benefits from a certain degree of autonomy. Thus, having access to a database and operating based on the instructions provided by the trader, the AI Agent may engage in conversations with users that go beyond the mere role of presenting a catalogue of offers tailored to the client’s needs.
IV.1. Negotiation Conducted by an AI Agent
As discussed earlier, AI Agents may assume more complex roles, such as negotiators or domain experts.
For example, AI Agents can be deployed on commercial platforms to allow users to negotiate product prices. Once the user identifies a product of interest, they may enter a dedicated negotiation interface. The AI Agent, following its programming, is aware that the product may be sold within a certain price range rather than at a fixed price.
When the user proposes a lower price, the AI Agent determines—based on preset thresholds—whether the sale can proceed.
Here, the matter becomes more nuanced because the user’s proposal modifies the original offer. The question arises: Can an AI Agent accept a counteroffer?
One could argue, by applying the legal principle a maiori ad minus, that since a person can delegate full authority to another via a mandate contract, then programming a system to accept an offer may constitute a more direct form of expressing consent.
Nevertheless, a more reliable approach, in my view, is found in Article 1.197(2) of the Civil Code, which defines a modified offer as a counteroffer:
“(2) The response of the offeree, expressed under paragraph (1), may, depending on the circumstances, be deemed a counteroffer.”
The key lies in the phrase “depending on the circumstances.” I propose that users should be informed that any modification to the trader’s offer shall not be deemed a counteroffer but a request for quotation.
Accordingly, the AI Agent’s response would constitute a new offer reflecting the trader’s updated instructions. Thus, only the user’s consent would be needed for contract formation, similar to the first scenario.
IV.2. Recommendations by an AI Agent
Where the user requests recommendations from the AI Agent, legal qualification is more straightforward.
For instance, a user may visit a wine merchant’s platform seeking the best wine pairing for beef. The AI Agent, accessing the merchant’s database or an external one, determines the appropriate wine.
If the user specifies, “I need a wine for grilled beef steak,” the AI Agent identifies a suitable match—say, Cabernet Sauvignon—and formulates an offer from the trader’s catalog, once again leading to the scenario outlined in Section III.
Ultimately, the AI Agent’s activity currently remains confined to presenting pre-established offers.
V. Conclusions
We must acknowledge that, at least for now, AI Agents cannot be considered distinct legal entities with autonomous will or decision-making capacity. They remain extensions of their human users—mere communication tools. AI Agents do not think, speak, or negotiate; they simply reproduce the intent of their programmers.
Their utility for businesses is evident, striking a balance between clients’ need for interaction and the commercial need for efficiency. However, the entirety of the AI Agent–client interaction is essentially a structured query of a commercial database and a user-friendly expression of the merchant’s offer.
This analysis can become more complex depending on the AI Agent’s sophistication, decision-making autonomy, learning capabilities, or deviation from instructions.
Still, I emphasize that legal analysis must avoid artificially inflating complexity or anthropomorphizing AI Agents. At present, they are intelligent tools for information processing and communication—tools that simulate human interaction but remain tools nonetheless.
Accordingly, a contract concluded through an AI Agent, as described above, merely represents the evolution of communication methods between businesses and consumers—a new chapter in online commerce.
[1] https://ec.europa.eu/eurostat/web/products-eurostat-news/w/ddn-20250123-3
[2] Interaction generated through the ChatGPT program following the instructions in the provided example.
[3] C. Hamangiu, I. Rosetti-Bălănescu, Al. Băicoianu, Tratat de drept civil român, vol. II, Ed. All Beck.
Article published by Juridice.ro: https://www.juridice.ro/781888/incheierea-contractelor-prin-intermediul-inteligentei-artificiale.html