You ask an AI agent to reduce your cloud costs. It delegates the task to another agent, which deletes an important archive.

Who authorized the deletion?

As AI agents begin making decisions, spending money, and delegating tasks to other agents, the question of responsibility becomes harder to answer.

Tracing which agent performed an action is one thing. Tracing who gave it authority, what limits applied, and whether those limits survived delegation is another.

In her latest article, Sasha Shilina explores this growing problem.

If an AI agent exceeds its authority, who should be held responsible: the agent's operator, its developer, or the person who gave the original instruction?

Read the full article: https://blog.humanode.io/who-authorized-this/