Who Is Responsible When an AI Agent Breaks the Rules

On October 30, 2026, an article was published about how companies use AI agents (programs that perform tasks on their own: transfer money, edit code, conduct sales) and who is responsible if such an agent breaks the rules. The note poses two questions: how to confirm the identity of each agent and who bears responsibility for its actions. There is no data in the source about specific incidents, companies, or court decisions.

The difference from previous discussions is that earlier the focus was mainly on chatbots that merely advise a person. Here the agent gets access to money, code, and sales and acts on its own. Therefore the question shifts from the quality of answers to accounting for actions: who exactly performed the operation and on whose instruction. This is a task about identification (linking an action to a specific agent and its owner), not about the quality of the model.

In practice, companies need a way to issue credentials to an agent, limit its rights, and keep an action log. Without this, it is impossible either to revoke access or to sort out a disputed case. Responsibility in this case depends on what is recorded in the contract and in the logs: who the owner of the agent is, what rights were granted to it, and who approved the action. The source does not provide examples of how this is arranged at specific companies, so the details remain open.

What remained unknown: how exactly companies confirm the identity of agents and what responsibility rules are already applied. The note contains no examples, no figures, and no references to regulators.