Styx built an entire concept album in 1983 around a man who escapes by hiding inside a machine, and gave the machine the album’s most famous line — a polite word of thanks for doing the dirty work nobody wanted. The joke, forty years on, is that we have built the machine and handed it the dirty work, and the thanks is about to be followed by an invoice.
If that still sounds like a problem for later, consider the version you have already lived through. Anyone who has been sent an electronic signature packet holding a dozen or more separate instruments has clicked through part of it on trust, applying a signature across documents nobody read in the minutes it actually took. The law settled long ago that you are bound by those. Agentic AI does not introduce the problem of assent without attention — it just removes the last human from the loop, and as we will see, the legal answer does not change much when it does.
Everything I have written about AI in this series has treated it as something that drafts. The next wave does not draft. It acts. Agents now place orders, accept terms, renew subscriptions, approve invoices, and click through the agreements nobody reads — without a human seeing the transaction until it appears on a statement. The question that follows is not philosophical. It is the oldest question in contract law: is my client bound?
The Problem Is Plain Enough
Start with what is actually being deployed, because the hypotheticals are already behind the market. Procurement agents that reorder inventory against a budget. Scheduling agents that book and pay for services. Vendor-management agents that accept renewal terms rather than let a license lapse. Customer-service agents that make commitments on a company’s behalf in a chat window a court will later read as a contract.
In each case a counterparty receives what looks like assent from your client’s system, on your client’s account, within your client’s apparent authority. The fact that no employee read the terms is, from where the counterparty sits, invisible — and as we will see, largely irrelevant.
Machines That Were Meant to Save Us
Here is the part most business owners find surprising: the law anticipated this, and it did not anticipate it in their favor.
The Uniform Electronic Transactions Act, adopted in Arizona and nearly every other state, defines an “electronic agent” as a program or automated means used independently to initiate an action or respond to electronic records, in whole or in part, without review by an individual. The federal E-SIGN Act carries parallel language. Those definitions were written decades ago for automated ordering systems, and they fit modern AI agents comfortably enough that no court has needed to stretch them.
The attribution rule is the one that matters. Under UETA, a contract may be formed by the interaction of electronic agents, and it binds the user even though no individual reviewed or was aware of the agents’ actions. The official commentary is blunt about the theory: when machines are involved, the required intention flows from the programming and the use of the machine. Translated into plain English, the law treats your agent’s assent as yours, because you chose to deploy it.
So the instinctive defense — nobody here agreed to this — is not just weak. It is the precise argument these statutes were written to foreclose.
The Man Behind the Mask
That leaves agency law, which is where the real fights will happen, and where there is almost no case law yet to guide anyone.
If your agent exceeds the authority you gave it — buys past the budget, accepts an indemnity you would never sign, renews a contract you intended to terminate — you are in a dispute about actual versus apparent authority, with a twist no reported decision has yet resolved. Apparent authority normally turns on what the principal’s conduct led the third party to believe. A counterparty transacting with your system, on your credentials, has a strong story. Your story — that the system misread its own instructions — requires proving what your configuration was at the moment of the transaction, which most companies cannot do, because nobody is logging it.
UETA does provide a narrow escape hatch for errors, and it rewards companies that built safeguards. Where a party has established a security procedure — a human approval step, a verification requirement — the error provisions can be available. Where no such procedure exists, the fallback is ordinary contract law: mistake, unconscionability, good faith. Those are real doctrines, but they are not where you want to be standing.
And one more thing to check before anyone relies on a vendor to make this right. The terms of service governing the agent itself will likely control the allocation of risk between your client and the developer, and those terms are not written in your client’s favor.
Secret, Secret
The useful work here is preventive, and the window is open now, while these terms are still negotiable and before anyone has a bad case to point at.
In contracts your client signs, define which agents may transact, set explicit dollar and scope limits, require human approval above a threshold, and require the counterparty to verify anything outside those limits. Say what happens when an agent exceeds authority — silence will be read against the party who deployed it. Build the audit trail on purpose: which agent, which version, which instruction, which approval, preserved in a form you could hand a court two years later.
On the other side of the deal, the diligence questions are the mirror image. Is the counterparty transacting through an agent? Who configured it, and who at that company signed off on the parameters? If you are the one trying to enforce, you want that record to exist too.
Thank You, Machine
The song is sung by the machine, explaining that it did what it was built to do and that the man inside is the one with a secret. That is the posture courts are likely to take. The agent will not be the defendant. The business that deployed it, and signed nothing, and knew nothing, will be — and the statutes already on the books say that is the right result.
The companies that come through this well will be the ones that decided in advance what their machines were allowed to promise, and wrote it down.
This post is part of an ongoing series on how AI is reshaping litigation and the businesses that live through it. Note for readers who like their law settled: there is very little case law here yet. That is the reason to deal with it in your contracts rather than in a courtroom.
A Word About Silver Cain
Silver Cain PLC represents businesses in complex commercial and real estate litigation in Arizona and beyond. A good deal of our work involves what a company’s own systems and documents committed it to before anyone in the room realized a commitment had been made. If the questions in this post are relevant to your business, or to the firms you retain, we are glad to have that conversation.
Leon Silver is an AV-rated trial lawyer at Silver Cain PLC, focused on commercial and real property disputes since 1989. Reach him at lsilver@silvercain.com.
