Agent-initiated dispute

A dispute in which a cardholder asks their bank to reverse a purchase their artificial intelligence (AI) agent made, saying the agent was not allowed to make it, bought the wrong thing, or went beyond what they asked for.

A person asks an agent for an inexpensive replacement filter, the agent buys one at double the price, and the person files a chargeback. Or someone who approved a purchase later blames their agent to get a refund while keeping the item.

A chargeback on a transaction an AI agent started under delegated authority. The case falls between existing card network categories: the person usually authorized the agent, but may not have authorized the specific purchase. The agent made the purchase, and the cardholder files the dispute.

These disputes fall into four groups:

• Agent error, where the agent misread an instruction or picked the wrong item.
• Scope breach, where the agent went past the person's limits on price, category, or merchant.
• Hijacked agent, where an attacker took control of the agent, for example through prompt injection.
• False claim, where the person wanted the purchase but blames the agent to get a refund.

As of November 2026, no card network has introduced a liability shift or dispute category specific to agent purchases, so these cases run through existing chargeback rules, although Visa's public rules began addressing agentic transactions in April 2026.

An agent-initiated dispute sits next to terms that are easy to mix up:

• Friendly fraud is a genuine cardholder disputing their own purchase, and the false "my agent did it" claim is a new version of it.
• Chargeback is the dispute process these cases go through.
• Agent mandate defines what the agent was allowed to buy, and is the main evidence.