Your AI agent, your liability

A federal appeals court just decided that when an AI agent logs into a site, the user did it, not the company that built the bot.

By ·

Sep 14, 2026

7 cards · 1m 50s · tap to begin

From · · · 1m 50s

Your AI agent, your liability

A federal appeals court just decided that when an AI agent logs into a site, the user did it, not the company that built the bot.

In brief

A federal appeals court just decided that when an AI agent logs into a site, the user did it, not the company that built the bot. Under the CFAA, an AI agent is a tool and the person directing it is the one doing the accessing, so site owners need contracts and other claims, not hacking law. Originally reported by…

Who pressed the button

Who pressed the button

When an autonomous agent logs into a password-protected site, the Ninth Circuit says the person who told it to go is responsible. Not the developer, and not the agent.

Amazon loses its injunction

A three-judge panel vacated a lower court order that had barred Perplexity from reaching password-protected Amazon accounts. The 21-page opinion found Amazon unlikely to win its Computer Fraud and Abuse Act claim.

How the fight started

Amazon warned Perplexity's CEO before Comet launched in 2025 that its AI could not touch the Amazon Store, then repeated it after launch. When Comet's Assistant started scraping, Amazon sued under the CFAA and California's CDAFA.

The CFAA is not a catch-all

The court warned that plaintiffs cannot stretch a computer-hacking statute into a general tool for policing online behavior they dislike.

“plaintiffs cannot bootstrap the CFAA into a general-purpose law for policing unwanted online conduct”

The user does the accessing

Access means entering a system or its files. On these facts, the panel said Perplexity is not the one entering Amazon's computers.

“It is the user who 'accesses' Amazon's computers, with the help of the Assistant to carry out specific acts on Amazon.com.”

An agent is a tool, not a person

Agentic AI Poses New Antitrust Risks, Legal Experts Warn

However capable the Assistant looks, the court refused to treat it as a legal person. And it drew a tight box around the ruling: no new regime for agentic AI, no view on tort liability.

“However advanced the Assistant currently is, it is a tool, not a person for statutory purposes”

Architecture becomes legal risk

Lawyers at ZwillGen read the opinion as a signal that how much autonomy an agent has, and how much the developer steers each interaction, will drive CFAA exposure. Site owners still have terms of service, contract, copyright, DMCA and tort claims.

The big picture

Under the CFAA, an AI agent is a tool and the person directing it is the one doing the accessing, so site owners need contracts and other claims, not hacking law.

Read the original on Pymnts

React

Sign in to react and comment.

Comments (0)

Life is short. Keep it sweet. Respect others' opinions and be kind!

    Recommended next

    More decks on ai agents and related topics.