Asenion AI Guardian
Asenion AI Guardian delivers Runtime Governance — the fourth of Asenion’s four capability areas, governance at the moment of action. Where Authoring turns rules into cited controls, Assessment is the system of record, and Testing produces evidence before and beyond deployment, Guardian is the layer your running agents and applications consult while they act.
For the whole lifecycle in one page — Authoring, Assessment, Testing, and Runtime Governance as one connected record — start with What is Asenion?.
What Guardian is
Asenion AI Guardian is Asenion’s runtime governance layer: an advisor and witness service. Your agents and applications consult it at the moment of a bounded action, and it surfaces the policies and guidance that apply. Every consultation produces a tamper-evident witness record: what was consulted, what guidance was returned, cryptographically chained so it can be verified afterward.
Guardian is reached programmatically, as an MCP service your integration calls. Integrating Asenion AI Guardian describes the tools and when to call each; Witness records describes the record it produces and how that record is verified.
In this section
| Page | What it covers |
|---|---|
| Witness records | What a witness record contains, how it is chained so it can be verified, and how a consultation and a later evaluation are paired |
| Integrating Asenion AI Guardian | The MCP service, its three tools, when to call each, a quick-start, and a worked example |
Where it fits the four stages
Authoring classifies each control by how it must be governed — “a documented obligation to assess, a measurable property to test, or a live condition an operating agent consults at the moment of action.” Runtime Governance is that third way. A control routed to runtime is one an operating agent consults through Guardian while it acts, rather than one a person assesses in a workflow or a test measures in advance.
That is what makes the four capability areas one system: the same source text that becomes a control can be assessed, tested, and consulted at runtime, and each of those leaves a record attached to the system it governs.
The consult, guidance, and witness-record loop
A consultation is a single call an agent or application makes before a bounded action:
- Consult. Your integration calls Guardian before the action, naming what it is about to do.
- Guidance. Guardian surfaces the policies and guidance that apply. It does not return a verdict.
- Witness record. The consultation produces a tamper-evident witness record — what was consulted, what guidance was returned — cryptographically chained so it can be verified afterward.
An agent’s response after the action can be evaluated in a second, paired call, producing a paired witness record linked to the original consultation. Both records, and the link between them, are described in Witness records.
Integrate in three steps
| Step | Call | What happens |
|---|---|---|
| Before the action | consult_policy | Guardian surfaces the applicable guidance and produces a witness record with an event_id |
| Your integration acts | — | Your code decides how to act on the guidance — proceed, change course, escalate to a person |
| After the response (optional) | evaluate_response | Guardian emits a paired witness record linked to the consultation |
The full walkthrough, including a worked example, is in Integrating Asenion AI Guardian.
Guidance, not verdicts
Guardian advises and witnesses; it does not act for you. This is the design position the rest of the section is built on, and it is stated two ways that mean the same thing:
Your integration decides how to act on the guidance — Guardian’s record proves what was surfaced, and when.
Guardian does not block or approve actions on its own. It surfaces applicable guidance and witnesses what happened. Decisions — human or automated — belong to your integration. That is a design position, not a limitation: a governance layer you can trust is one that shows its work rather than one that silently acts.
Guardian sits at the moment of a bounded action and produces governance evidence. What to do with the guidance — proceed, change course, escalate to a person — is your integration’s to determine, and the witness record proves what guidance was on the table when that determination was made.
Where to go next
- Witness records — what a witness record contains, how it is chained so it can be verified, and how a consultation and a later evaluation are paired.
- Integrating Asenion AI Guardian — the MCP service, its three tools, when to call each, and what your integration does with the guidance.
- What is Asenion? — the four capability areas as one connected record.
- Trust & Security — the evidence model, for a security or compliance reviewer.
- Glossary — Guardian, consultation, and witness record defined in one place.