Agent Authority
A legal reference by Obehi Ota

The law of software acting for companies

When an AI agent acts for a company in its dealings with another business, is the company bound, and who pays if it gets it wrong?

Often, yes. Whether the company is bound depends on what the parties agreed to and whether the agent followed the steps for making a deal. Who bears a loss can also turn on the contract, what the other side knew and whether a mistake gives the company a way out.

This site helps companies decide what their AI agents can commit them to, enforce those limits in their systems, and keep records showing what they authorized and what their agents actually did.

Where the risk sits

What the system allows isn’t what the company approved

A company’s purchasing policy lets its agent spend up to $10,000 an order. Its ordering system lets it submit larger orders. The card issuer sets a separate $100,000 limit per payment. Move the slider.

$42,000
Company approved: up to $10,000Card limit: $100,000 per payment
Above the policy limit, within the card limit

The agent can submit the order, and the amount fits the card limit. The seller decides whether to accept the order. Check what the contract required for this order, whether the seller knew about the $10,000 limit, and whether the company used the goods after learning what happened.

Read the guide: Technical Permission vs. Legal Authority →

Who’s behind it

Obehi Ota

A technology transactions attorney with almost two decades of legal experience, including Simpson Thacher, IBM and Wilson Sonsini, and degrees in computer information systems and information science. Licensed in New York and California. The guides cite their sources throughout. About the site →