Agents can act. They cannot yet prove.
Why the missing piece in the AI economy is accountability, and how HANRIA is designed to build it. A response to the essay Principals.
Blog
Essays and explainers on accountable agents.
Notes from HANRIA on what it takes for an AI agent to be trusted with real work.
Essays
Governed Habitats: What the Agent-Habitat Literature Asks of a Harness, and Where HANRIA Stands
An assessment of the HANRIA harness against the agent-habitat literature: what aligns, what does not yet, and what external validation would take.
Explainers
One page at a time, in the order they were written. Each page keeps its own URL.
A skill can check an agent action. Only a runtime can stop one.
A skill can check a proposed agent action against a written mandate and record permit, deny, or escalate. Stopping the action requires exclusive control of its credential or tool.
What is an agent mandate, and why a permission system will not catch authority drift?
An agent mandate is a written grant of which operations, targets, counterparties, and amounts an agent may attempt, when it expires, and which kinds always need a person. A permission system will not catch authority drift.
How to write an agent mandate: denials first, expiry, requires_human
How to write an agent mandate: denials first, expiry, requires_human.
What is authority drift when every individual agent check still passes?
What is authority drift when every individual agent check still passes?
What should an agent do on permit, deny, escalate, or error?
What should an agent do on permit, deny, escalate, or error?
Worked mandate templates: ticket read, file prefix, refund ceiling
Worked mandate templates: ticket read, file prefix, refund ceiling.
What a hash-chained local decision log is, and is not
What a hash-chained local decision log is, and is not.
What is the difference between an agent skill and an enforcement runtime?
What is the difference between an agent skill and an enforcement runtime?
What does fail-closed mean for an agent mandate?
What does fail-closed mean for an agent mandate?
When must an agent escalate instead of acting?
When must an agent escalate instead of acting?
Developer guides
Developer guides for pre-action checks and agent mandates
LangGraph, OpenAI Agents SDK and Claude Agent SDK pre-action checks, plus writing and validating a mandate.