Who is liable for what? The liability map of the agent economy
Last updated: · Leutrim Miftaraj
Four parties, four spheres of risk: the person granting the mandate is liable for actions within the mandate (it is their declared will); the agent operator for malfunction outside the instruction; the verifier for acceptance without checking; the infrastructure for wrong verification answers.
Within the mandate
What the agent does within the frame is attributed — which is why precise limits are the principal's self-protection, not distrust of the agent. The template gallery supplies default limits per action type and risk level.
Outside the mandate
If the agent oversteps, the mandate does not apply — the verifier that checked is protected by the receipt; the damage lies in the agent operator's sphere. Without a check the picture reverses: whoever accepts blindly shares the risk.
The infrastructure duty
Fail-closed as a principle: if the status cannot be checked, the answer is deny(unavailable) — never a false allow from the cache. Availability is negotiable; a false yes is not.
Key takeaway
Clear spheres instead of passing the buck: the mandate places liability where control lies.
Try it yourself
Issue your first agent mandate and check your first decision with its receipt — free, in a few minutes.