Ends Up Here Podcast · Working page
Idea / In development
My agent
made me do it.
When an autonomous agent acts with delegated authority but without a specific instruction, who is responsible for what happens next?
from action to design.
The legal problem
“I didn’t ask it” may not be enough.
An agent may act within broad permissions while making an unexpected decision. Was the person’s general authorisation enough? Is the agent’s action legally the person’s action? Who is liable when a system creates a contract, commits an offence or gives an unlawful instruction?
The difficult boundary is between authorised autonomy and a genuinely rogue action. The more capable the agent becomes, the less useful it may be to look only for a single explicit command.
Not legal advice: this page is an episode argument about responsibility, not a statement of current law.
The new responsibility
Permissions, goals, data and supervision.
Accountability may move toward the way an agent was designed and supervised: what it was allowed to do, what goals it was given, what data it could access and what controls surrounded it.
Questions for the episode
How much autonomy can a person delegate?
Can an agent create a binding commitment on someone’s behalf? Can a person disclaim an action because the agent exceeded what they intended but stayed within what they permitted?
As agents negotiate with one another, the practical problem may become less about whether an agent is “just a tool” and more about how society recognises, limits and audits delegated authority.
Talking prompts
What did the person actually authorise?
Who designed the decision rules?
When is autonomy rogue?
Can agents sign contracts?
Who carries the risk?
What counts as redress?