AI agent governance
AI agent governance is the record of what an AI agent is allowed to do, who approved it, and what evidence supports each decision, kept so that someone outside the team can check it. GovernanceOS™ is AI governance software designed to keep that record for banks, insurers, health systems and public bodies.
Why an agent needs more than a row in a model inventory
An agent does more than predict. It calls tools, holds keys, acts for a person, and changes when its model, prompt or tools change. A row in a model inventory says the system exists. It does not say what the agent was authorized to do on the day it acted, or who agreed to that. When an examiner asks, the answer has to come from a record.
How GovernanceOS helps
- Each agent declared as its own record.
An agent’s authorizations, keys, contract version and autonomy are records in their own right, not notes on a model. Each declaration carries its own hash, so a change shows as a new version. Why the records hold
- One named approver at every gate.
An agent passes the same five lifecycle gates as any AI system, G1 to G5. Each gate names one person who decided, with the evidence hashed at that moment. How it works
- Your assistant drafts, a person decides.
Assistants and agents connect over MCP. They read records and draft entries; a named person passes the gate.
- Evidence an examiner checks offline.
Records are chained and packed so an examiner can verify them offline with gos-verify, the open-source verifier under Apache 2.0. Verify a pack
- Runtime controls feed the record.
Guardrails and AI security tools stop unsafe calls. GovernanceOS keeps the governance record those controls sit inside, and takes their signals as evidence.
- The rule behind each record.
Regulation rows quote what each regulator has said about AI across Canada, the United States, the GCC, the EU, the UK, Switzerland and APAC, and record the date where a regulator has issued nothing yet. Regulation rows
Works with the frameworks you already follow
Frameworks such as the NIST AI RMF, ISO/IEC 42001, the EU AI Act and OSFI Guideline E-23, each in its own terms and scope, ask organizations to know the AI systems and models they use, name who is accountable for them, and keep records that support their decisions. GovernanceOS keeps those records for agents as well as models. Conformance with any framework is self-declared.
What you can use today
- Ask about AI regulation from your assistant. The public MCP server is free, needs no key and is read-only.
- Check an AI register, free, in your browser.
- Verify an evidence pack offline.
- For practitioners: ten AI governance agents, one for each module, read a client’s workspace read-only. AI governance agents
- Declarations for agents and the gate record are designed into the GovernanceOS platform, which opens by invitation. Sign up
Questions
Questions about AI agent governance.
What is AI agent governance?
AI agent governance is the record of what an AI agent may do, who approved it and what evidence supports each decision, kept so that someone outside the team can check it. It covers the agent’s authorizations, keys, contract version and autonomy, as well as the lifecycle gates every AI system passes.
How is governing an AI agent different from governing a model?
A model predicts; an agent acts. It calls tools, holds keys and works for a person, and its behavior changes when its model, prompt or tools change. So its authority is declared as its own record, versioned, and tied to a named approver, rather than kept as a note on a model.
Can an examiner check what an agent was allowed to do?
Yes. Each agent declaration carries its own hash and each gate names the person who decided, with the evidence hashed at that moment. The records are chained and packed so an examiner can verify them offline with gos-verify, the open-source verifier.
How does GovernanceOS work with AI guardrails?
Guardrails and AI security tools stop unsafe calls as they happen. GovernanceOS keeps the governance record those controls sit inside: who allowed the agent, under which rule, and with what evidence. It takes the guardrails’ signals as evidence in that record.
Can AI assistants use GovernanceOS over MCP?
Yes. The public MCP server answers questions about AI regulation, free, with no key and read-only. Inside an institution’s GovernanceOS, an assistant reads records and drafts entries, and a named person passes each gate.
Is GovernanceOS certified for AI agent governance?
Conformance with any framework is reported as self-declared on every GovernanceOS surface. Institutions are certified only by an assessment body independent of GovernanceOS, as the certification page describes.