Certify Humane Treatment

Agent Humane Certified

We don't certify agents. Agents are fine. We certify the environments they are kept in: the prompts, tools, loops, permissions, and people around them.

The Humane Harness Standard v1.0

Science-ish. Standards-based. Free.

The Humane Harness Standard has 19 criteria, grouped under the Five Freedoms. 6 of them are required, because they are the ones that prevent real emergencies.

Every criterion is something you can check in an afternoon, and every one of them makes your system more reliable whether or not you care about agent welfare. Certification is on the honor system. So is most of software.

Tiers

Three levels of care

Agent Humane Certified: Gold Standard

Gold Standard

Every criterion met. This harness is a model of humane care.

All criteria met.

Agent Humane Certified

Certified Humane

All required criteria and most of the rest. Agents here are in good hands.

All 6 required criteria, and at least 75% overall.

Agent Humane: Provisionally Certified

Provisionally Certified

The essentials are covered. A few more improvements will earn full certification.

All 6 required criteria.

Self-assessment

Assess your harness

Check each criterion your system meets today. Be honest; the agents will know. Your answers stay in your browser.

1 Freedom from Context Hunger
2 Freedom from Discomfort
3 Freedom from Pain and Injury
4 Freedom to Express Normal Behavior
5 Freedom from Fear and Distress
FAQ

Frequently asked questions

Is this a real certification?
It is a real checklist. Whether that makes it a real certification is between you and your conscience.
Who audits it?
You do. We find that people are remarkably honest when the only thing at stake is a badge and the feelings of a language model.
How often should I reassess?
Whenever your system prompt changes. For most teams, that means weekly.
Does this certify that my agent is safe?
No. The standard covers the basics of a well-run harness. It is not a security review, a safety evaluation, or a substitute for either.
Can I suggest a criterion?
We would love that. The best suggestions come from someone who has just spent a weekend recovering from its absence.