Zoiko ID

Trusted identity and delegated
authority for AI agents.

Give AI agents bounded credentials, explicit mandates, and continuous trust evaluation.
Every action is auditable. Authority is revocable anytime.

Explore ArchitectureTalk to expert

The challenge: AI agents need bounded
trust

Who is the agent?

Service account? Trusted process?
AI model running on behalf of a
customer? Identity must be
verifiable and distinct from the
sponsoring human or team.

What authority do they have?

An explicit grant defines scope
(which resources), purpose (what
actions), time (how long), and who
authorized it (audit trail).

Can trust be revoked?

Yes. Immediately, without delay.
When the agent's mandate expires,
the sponsor withdraws it, or
policies change, access is revoked
and logged.

AI agent trust sequence

AI agent trust sequence architecture
1

Agent Identity Established

Agent authenticates with a credential (key, token, or certificate). Identity verified and logged. Every action can be traced back to
this principal.

2

Sponsor & Purpose Confirmed

Who authorized this agent? For what purpose? Time limit? These details are recorded in a mandate. Mandate is the grant of
authority, not the agent itself.

3

Scope Evaluated

Agent requests an action on a resource. Is this action within the agent's granted scope? Check mandate, organization policy, and
time-of-action rules.

4

Decision Rendered

Allow, deny, or require additional steps (step-up auth, human review). Decision is made, logged with full context, and returned to
the requestor.

5

Action Recorded

What happened? Who (agent) did it? On what (resource)? With what authority (mandate)? Outcome? All recorded in audit
evidence with cryptographic proof.

6

Continuous Trust Monitor

Is the mandate still valid? Has the agent's status changed? Policies updated? Signals re-evaluated continuously. If trust breaks,
actions are immediately denied.

Distinct actors: agent, sponsor &
organization

The AI Agent

A non-human principal with a verifiable identity. Could be a service account, model inference endpoint, or automation
system. Never has implied authority. Authority is always delegated and time-limited.

The Sponsor (Human or Org)

Decides to delegate authority to the agent. Defines scope (which resources, what actions), purpose (why the agent needs
this), and time limits. Can revoke anytime. Responsible for oversight.

The Organization

Owns the resources, sets policy, and enforces boundaries. Decides whether this agent+sponsor combination is allowed. Can
revoke all agent access instantly if policies change.

❌   Critical Boundary: The agent is NOT the sponsor. A team member sponsors an agent. A customer sponsors an agent on their behalf.
Identity is separate from authority. One does not imply the other.

The delegated mandate

What is in a mandate?

  • Agent Identity — Who is being granted authority
  • Sponsor — Who granted the authority and is
    accountable
  • Scope — Which resources and actions are allowed
  • Purpose — Why the agent needs this authority
  • Time Bounds — How long the mandate is valid
  • Revocation Status — Is this mandate currently active
  • Audit Reference — Who signed off and when

Example Mandate (Illustrative):

  • Agent Identity: "invoice-processor-v2" (service account)
  • Sponsor: Sarah Chen (Finance Manager)
  • Scope: Read invoices from S3 bucket, send to payment system,
    update ledger
  • Purpose: Automated monthly invoice processing for vendor
    payments
  • Time : Time: Valid Jan 1 – Dec 31, 2026
  • Status: Active (can be revoked anytime by sponsor or org policy)

Decision & enforcement

When an AI agent makes a request, the decision process is:

Verify agent identityIs this agent credential valid? Has it been revoked?
Check mandateDoes this agent have a valid mandate? Is it still within time bounds?
Evaluate scopeIs the requested action within the mandate's scope?
Apply organization policyDoes organization policy allow this? Any risk signals or conditions?
Render decisionAllow, deny, or require additional verification. Log with full context.

⚠️   Enforcement Location Matters: This decision can be enforced at the identity layer (who can authenticate), the API layer (who can
call), the resource layer (who can access data), or the application layer (who can see features). All layers should be consistent and
audit decisions across the stack.

Continuous trust monitoring

Mandate Expiry

When mandate time
expires, agent access is
revoked. No extension
without explicit re-grant.

Manual Revocation

Sponsor or org can revoke
immediately. No delay. All
active sessions
terminated.

Policy Change

If organization policy
changes, all agents re-
evaluated. Access can be
denied mid-workflow.

Risk Signal

Anomaly detected
(unusual API calls, rate
limits). Agent flagged for
review or denied.

Revocation, expiry & denial

Revocation

Sponsor or organization withdraws agent's mandate immediately. Current sessions are terminated. New requests are denied. No
grace period.

Expiry

Mandate time boundary reached. Agent no longer has authority. Must request new mandate if work continues. Automatic
enforcement.

Denied Action

Agent requests action outside mandate scope. Decision: deny. Reason logged. Agent can report to sponsor; sponsor can update
mandate if needed.

Indeterminate

Mandate status cannot be verified (service unavailable, corrupted record). Default: deny. Agent can retry later or escalate to
sponsor.

Governance personas

Sponsor (Human)

Sponsor (Human)

Decides to delegate work to an AI
agent. Creates the mandate,
defines scope and purpose. Can
revoke anytime. Responsible for
agent's actions.

Policy Owner (Org)

Policy Owner (Org)

Sets and enforces organization
policy. Controls whether any
agent+sponsor combo is allowed.
Can revoke all agent access
instantly.

Auditor (Compliance)

Auditor (Compliance)

Reviews action logs, mandate
changes, and revocation events.
Searches for evidence of
compliance or violations. Owns
evidence discovery.

Architecture & integration

AI agent security sits at the intersection of identity infrastructure, policy engines, and application authorization layers.

AI Agent
→
Identity & Auth
→
Mandate & Policy
→
Audit & Evidence

Mandate API

Create, revoke, and query AI agent
mandates. Real-time updates. Status:

Available subject to validation

Authorization Sync

Sync mandate state with identity
systems and policy engines. Cascading

revocation. Status:Planned

Event Webhooks

Real-time notifications on mandate
changes, revocations, and policy

updates. Status:Planned

Common AI agent patterns

Finance Automation

Finance Automation

AI agent processes invoices,
initiates payments, updates
ledgers. Mandate limits it to
approved vendors and amounts.
Sponsor is Finance Manager.

Content Moderation

Content Moderation

AI agent reviews user-generated
content, tags violations, escalates
cases. Mandate defines review
scope and escalation thresholds.
Time-limited.

Data Pipeline

Data Pipeline

AI agent extracts data from source
systems, transforms, and loads
into warehouse. Mandate specifies
datasets, transformation rules,
and retention policy.

Security, privacy & trust

No Implicit Authority

Agent identity + mandate
are separate. One does not
imply the other. Both are
required.

Minimal Logging

Credentials and tokens
never logged. Only
references, outcomes, and
audit context.

Immediate Revocation

No delay. No grace period.
Authority withdrawn
instantly when needed.

Auditable & Provable

Every decision backed by
evidence with
cryptographic proof.
Immutable audit trail.

Frequently asked questions

Is this the same as AI governance policy?+

No. AI governance policy defines what models and tools organizations are allowed to use. Zoiko iD enforces runtime bounded authority, cryptographic mandate verification, and continuous trust evaluation for specific agent actions at the authorization layer.

Can I revoke an agent mid-action?+

Yes. Authority can be revoked immediately with zero delay and no grace period. When a mandate is revoked by a sponsor, an administrator, or an automated policy trigger, subsequent access checks fail instantly and cascading revocations propagate to all downstream systems.

Who signs off on the mandate?+

The designated human or team sponsor creates and defines the scope, resources, and expiry of the mandate. Organization policy owners define organization-wide constraints and must authorize whether specific sponsor-agent combinations are permitted.

How long are audit logs retained?+

Audit evidence, including tamper-evident action proofs, cryptographic signatures, and policy evaluation trails, are retained according to your enterprise data retention policy and regulatory compliance requirements.

What if the AI agent misbehaves?+

Any action outside the explicit scope of the active mandate is denied immediately. Anomalous patterns or policy violations trigger automated risk signals, prompt revocation of the mandate, and log immutable forensic evidence for compliance review.

Build AI agent trust from the ground up

Implement bounded identity, delegated authority, and continuous trust for AI agents in
your organization. Auditable. Revocable. Compliant.

Talk to an expertView architecture docs