Responsibility
Who built, operates, sponsors, and remains accountable for the agent?
Agree on what people should be able to count on.
The Open Agent Trust Standard is intended to bring agent builders, ministries, institutions, and other participants into a transparent space where they can develop shared expectations for agents people may choose to use.
Who this is for
People who build, evaluate, choose, operate, or are affected by agents.
The work welcomes agent builders and operators, ministries and institutions considering agent use, domain specialists, evaluators, contributors, and affected communities.
They are invited to help shape the foundation. Their inclusion here does not imply adoption, endorsement, or an institutional relationship with Empty Jars.
Why this matters
Trust should not depend only on a sales claim.
As more AI agents become available, ministries and institutions may be approached by different builders offering different promises, terminology, safeguards, and levels of transparency.
The aim is not to create another place where builders compete to “sell their agents.” It is to develop common ground where builders and potential users can agree on basic requirements, make relevant evidence visible, and learn from one another.
Ministries and institutions should be able to help shape those expectations—not merely receive a product after the important decisions have already been made.
The practical question
If I use this agent, what can I reasonably count on?
A useful standard should make important information understandable before a person or organization decides to rely on an agent.
Purpose and limits
What is the agent intended to do, what should it not do, and where is human judgment still required?
Evidence
Which claims have been declared, checked, tested, or independently reviewed—and by whom?
Change and recourse
How can people detect important changes, stop access, report problems, and check whether trust has been suspended or withdrawn?
Standards and certification
Shared requirements can support clear, limited assurance.
The standard would define agreed expectations and evidence. A future certification process could then show that an agent was assessed against a named version and scope.
Certification should communicate specific, evidence-based confidence within that named scope—not promise that every answer will be correct, that no harm can occur, or that an agent is suitable for every purpose. A certificate would not be institutional or denominational endorsement.
Who may assess agents, how independence is protected, how long certification remains valid, and how suspension or revocation works are questions for a future team to resolve.
Questions still open
The details should be developed together.
“Open” currently describes a publicly available specification and development process. It does not decide whether an agent's own source code must be openly licensed.
- Which basic requirements should apply to every participating agent?
- Which expectations should change according to purpose and risk?
- What evidence would make an agent's claims credible?
- Must agents be open source, or can private implementations provide enough transparency for meaningful assessment?
- Who may assess or certify an agent, and how should conflicts of interest be controlled?
- Which changes should require reassessment?
- How should complaints, incidents, suspension, appeals, and revocation work?
A public foundation—not an endorsement program.
No agent is currently certified, approved for deployment, or endorsed by Empty Jars through this project. The standard is still being established.