Skip to content
Technology, Privacy & AI

Agentic Commerce: Contracting Through AI Agents

Agentic commerce, where AI agents shop, negotiate and pay for users, fits partly within existing Canadian law. Electronic commerce statutes allow contracts formed by electronic agents and protect individuals who make errors with a merchant's agent, but attribution, payment authorization, consumer protection and liability allocation still depend heavily on contracts and platform terms.

AI shopping assistant completing a checkout, illustrating agentic commerce and contracting through AI agents

Key takeaways

  • Ontario's Electronic Commerce Act allows contracts formed by interacting electronic agents.
  • Individuals can avoid certain transactions with a merchant's agent after a material error.
  • Attribution of an agent's acts depends on the authority given and the terms involved.
  • Agent spending authority should be limited, logged and revocable.
  • Consumer disclosures and the Competition Act's drip pricing rule still apply.
  • Contracts must allocate liability among users, agent providers, merchants and platforms.

What agentic commerce is

Agentic commerce describes transactions in which AI agents, rather than people clicking through checkout pages, search for products, compare offers, negotiate terms, place orders and initiate payments on a user's behalf. Consumer assistants that book travel or reorder groceries, procurement agents that request quotes and issue purchase orders, and merchant-side agents that answer, upsell and close sales are all examples. The technology raises a familiar legal question in a new form: when software acts, who is bound, who pays, and who bears the loss when something goes wrong?

AI shopping assistant completing a checkout, illustrating agentic commerce and contracting through AI agents
In agentic commerce, AI agents act for buyers and sellers, so contract formation, payment authority and liability must be allocated in advance.

Canadian law already provides partial answers through electronic commerce statutes, contract law, payment rules and consumer protection. This insight explains how these frameworks apply to agentic commerce and how businesses can allocate risk with platforms, payment providers and users. It is general information, not legal advice.

Can an AI agent form a contract?

Canadian provinces have electronic commerce statutes based on the Uniform Electronic Commerce Act, which anticipated automated transactions decades before generative AI. Ontario's Electronic Commerce Act, 2000 defines an electronic agent as a computer program or other electronic means used to initiate an act or respond to electronic documents or acts, in whole or in part, without review by an individual at the time of the response or act.

Section 19 provides that an offer, acceptance or other matter material to forming or operating a contract may be expressed by electronic information or by an act intended to result in electronic communication, such as clicking an icon or speaking, and that a contract is not invalid only because it is in electronic form. Section 20 provides that a contract may be formed by the interaction of an electronic agent and an individual, or by the interaction of electronic agents.

These provisions mean that a contract is not unenforceable merely because no human reviewed it at the moment of formation. But they do not answer every question. Ordinary contract principles still determine whether the parties intended to be bound, what terms were agreed and whose agent acted. When an AI agent behaves unpredictably, those questions become harder.

Attribution: whose act is it?

The central question in agentic commerce is attribution: is the agent's action attributable to the user who deployed it? Electronic commerce statutes and general principles point toward attributing an electronic agent's acts to the person who uses it to act on their behalf, much as a business is responsible for the information its website chatbot provides. But users will argue that an AI agent acting outside their instructions, or misinterpreting them, should not bind them. The answer is likely to depend on the authority the user gave the agent, the configuration and limits set, what the counterparty knew, and the terms of the platforms involved. See our insight on AI chatbots and liability.

IssueKey questionWhere answers come from
FormationWas a contract formed by agent interaction?Electronic commerce statutes, contract law
AttributionIs the user bound by what the agent did?Authority given, platform terms, counterparty knowledge
MistakesCan an individual undo an error made in dealing with an agent?Electronic commerce statutes, mistake doctrine
Payment authorizationWas the payment authorized, and who bears unauthorized loss?Card network and account terms, payments regulation
Consumer protectionWere required disclosures and cancellation rights respected?Provincial consumer statutes, Competition Act
Liability allocationWho bears loss among user, agent provider, merchant and platform?Contracts and platform terms

Mistakes in transactions with electronic agents

Electronic commerce statutes contain a specific error rule for individuals dealing with another person's electronic agent. Under section 21 of Ontario's Act, an electronic transaction between an individual and another person's electronic agent is not enforceable by the other person if the individual made a material error, the electronic agent gave no opportunity to prevent or correct the error, the individual promptly notifies the other person on becoming aware of it, and, where consideration was received, the individual returns or destroys it as instructed and does not materially benefit.

This rule protects individuals transacting with a merchant's automated systems, and it encourages merchants to build confirmation and correction steps into checkout flows. It does not squarely address the case where the individual's own AI agent made the error. Merchants serving agentic commerce traffic should consider keeping confirmation steps that agents must pass, and users' agent providers should give users chances to review significant transactions. The two approaches together reduce disputes.

Payment authorization

Payments are where agentic commerce meets existing risk controls. Card network rules, issuer agreements and bank account terms allocate losses for unauthorized transactions, typically favouring cardholders who did not authorize a charge. When a user gives an AI agent standing authority to spend, it becomes harder to say a transaction was unauthorized, and disputes will turn on the scope of the authority and evidence of what the user approved. Payment networks and providers are developing mechanisms for agent-initiated payments, such as tokenized credentials tied to specific agents, spending limits, merchant restrictions and step-up authentication for higher-risk transactions.

Businesses that hold funds or perform payment functions for agentic commerce should also consider payments regulation. The Retail Payment Activities Act regulates payment service providers that perform payment functions as a service or business activity not incidental to another activity, with registration, operational risk management, incident response and safeguarding of funds requirements supervised by the Bank of Canada. An agent platform that moves or holds users' funds may need to assess whether it falls within the regime. See our fintech and payments practice.

Practical point

Design agent spending authority like a corporate signing policy: set limits by amount, merchant category and frequency; require user confirmation above thresholds; log every instruction and approval; and make it easy for users to revoke authority. Clear authority records are the best evidence in a payment dispute.

Consumer protection

Provincial consumer protection statutes impose disclosure requirements, cancellation rights and rules on internet agreements and recurring charges. These protections were written with a human consumer in mind: the consumer must be shown information, given an express opportunity to accept or decline, and provided with a copy of the agreement. When an AI agent transacts for a consumer, merchants must still deliver required disclosures and copies, typically to the consumer through the agent's channel, and should not assume that an agent's acceptance waives statutory rights. Consumer protection law will generally be interpreted to protect the consumer, not the merchant, in cases of doubt.

Federal competition law also applies. The Competition Act treats representations of prices that are not attainable because of fixed obligatory charges or fees, known as drip pricing, as false or misleading. Merchants presenting prices to AI agents should ensure that the price an agent sees is the full price a consumer will pay, since agents may compare and select offers based on headline figures. Agent providers that rank or recommend offers should avoid undisclosed paid placement that could mislead users.

Privacy and data

AI agents act on detailed personal information: preferences, payment credentials, addresses, purchase histories and sometimes health or financial data. Under the Personal Information Protection and Electronic Documents Act, agent providers and merchants must identify purposes, obtain meaningful consent, limit collection and safeguard information. Merchants receiving orders from agents should consider what information they actually need and how it is protected, and agent providers should be transparent about what they share with merchants and platforms.

Allocating liability with platforms

Because the law is unsettled, contracts and platform terms will do most of the work in agentic commerce. Key allocation points include the scope of authority users grant to agents and how it is evidenced; the agent provider's responsibility for errors, hallucinated terms and actions outside instructions; merchant obligations to honour or reverse agent transactions; responsibility for fraud, including agents compromised through prompt injection or credential theft; dispute and chargeback processes; and limitations of liability and indemnities among the parties. Merchants should review whether their terms of service permit automated purchasing and on what conditions, and whether they want to block, allow or specifically support agent traffic.

For business-to-business procurement agents, master agreements can address agent use expressly: which agents may submit orders, what authentication applies, which order types require human confirmation and how errors are corrected. See our insight on AI vendor contracts.

Quebec and other provinces

The analysis varies by province. Most common law provinces have electronic commerce legislation modelled on the uniform act, with similar provisions on electronic agents and errors, though wording and section numbers differ. Quebec's framework for information technology and its civil law of obligations, including rules on consent, mandate and the consumer protection regime, apply to transactions involving Quebec consumers and businesses. Businesses offering agentic commerce services nationally should map the rules in each province where their users and merchants are located rather than assuming that the Ontario approach applies everywhere.

Agency law and the user's AI agent

The word "agent" is suggestive, but an AI system is not a legal person and cannot itself be an agent in the legal sense. The more useful analysis treats the AI as a tool used by the user, or by the provider acting for the user, with legal effects flowing from the authority the user conferred and the representations made to counterparties. Where a provider operates the AI on the user's behalf, the provider may itself act as the user's agent or mandatary, with duties to follow instructions and act with care. Agentic commerce providers should therefore define in their terms what they do for users, what they do not do, and what standard of care applies.

Security risks

Agents that read web content and act on it are vulnerable to prompt injection, where malicious content on a page instructs the agent to do something the user did not intend, such as buying from a fraudulent merchant or disclosing payment details. Credential theft and impersonation of agents are other risks. These security issues are also legal issues, because they will influence how courts and payment networks allocate losses. Reasonable security measures, monitoring and incident response are likely to be expected of agent providers and merchants alike.

Preparing for agentic commerce

Merchants, platforms and payment providers can prepare now. Review terms of service for automated access and purchasing. Keep checkout confirmation and error correction steps that work for both people and agents. Present full prices and required disclosures in machine-readable form. Implement authentication and spending controls for agent-initiated payments. Log agent interactions for dispute resolution. Update privacy notices and data sharing arrangements. And negotiate clear liability allocation with agent providers and platform partners as integrations are built.

How TRW & Co helps

We advise merchants, AI agent developers, platforms and payment businesses on agentic commerce, including contract formation and terms of service, payment authority and payments regulation, consumer protection and pricing rules, privacy and liability allocation. See our artificial intelligence practice or contact us.

Frequently asked questions

Is a contract formed by an AI agent enforceable?

Generally it can be. Ontario's Electronic Commerce Act provides that a contract may be formed by the interaction of an electronic agent and an individual or between electronic agents, and that a contract is not invalid only because it is electronic. Ordinary contract principles still apply.

What if the user's AI agent makes a mistake?

The statutory error rule protects individuals dealing with another person's electronic agent where no opportunity to correct the error was given. Errors by the user's own agent are less clearly covered, so authority limits and confirmation steps matter.

Who bears losses from unauthorized agent payments?

It depends on card network rules, account terms and the scope of authority the user gave the agent. Clear limits, logs and authentication make disputes easier to resolve.

Do consumer protection laws apply when an agent buys for a consumer?

Yes. Merchants must still provide required disclosures, copies and cancellation rights, and the Competition Act's rules on misleading prices, including drip pricing, continue to apply.

Could an agent platform be a regulated payment service provider?

Possibly, if it performs payment functions such as holding or transferring funds as a business activity. The Retail Payment Activities Act should be assessed.

Sources

  1. Electronic Commerce Act, 2000, S.O. 2000, c. 17 (Ontario e-Laws)
  2. Competition Act, s. 74.01 (Justice Laws)
  3. Retail Payment Activities Act (Justice Laws)
  4. Personal Information Protection and Electronic Documents Act (Justice Laws)

General information only, not legal advice. References to statutes, courts and authorities describe published law and practice at the date shown.