Guard

Guard is the governance built into Cilila and Figena. Every access is scoped to a role, every important action starts as a draft someone approves, and every step leaves a signed record you can check.

Why Guard

AI that can act on business data has to stay inside the same limits as the people it works for. Guard keeps it there: it decides who and what may read or change a record, holds important actions for approval, checks data before it is sent to a model, and keeps a record that cannot be quietly rewritten.

What Guard does

  1. Scoped access

    Role-based access decides who can see and change what, for people and for AI alike.

    • Roles and access profiles with defined scopes.
    • Access changes reviewed as changesets.
    • Agents and outside tools given their own limited scopes.
  2. Checked before it reaches a model

    Each request to an AI model is classified first, and restricted data is held back.

    • Requests checked for sensitive patterns such as card numbers, identifiers and medical terms.
    • Restricted data stays out of model requests unless policy allows it.
    • A record of what data was sent to AI.
  3. Drafts and approvals

    Important actions by AI start as drafts, and a person approves them before anything changes.

    • One inbox for drafts waiting on a decision.
    • Checks before an action runs, and approval before an agent is released.
    • AI writes only through the governed path.
  4. A record you can trust

    Every access, action and approval is written to an audit trail that is signed and hard to alter.

    • Signed, hash-linked audit events, checked for tampering.
    • Sensitive fields redacted in audit logs and payroll views.
    • Scheduled evidence packs you can hand to a reviewer.

How Guard governs Cilila

  • Same limits as the user

    Cilila only sees and acts on what the person it works for is allowed to.

  • Checked before it runs

    Agent actions go through OpenPort, which checks scope and payload before anything is written.

  • Visible afterwards

    Admins can see what AI was asked, what it did and what was approved.

Where Guard helps

Payroll and HR
Pay and personal details shown only to the roles that need them, with redaction elsewhere.
Finance changes
Changes proposed by AI in finance wait as drafts until a person approves them.
Restricted data
Restricted records stay with approved accounts and out of AI requests.
Reviews and audits
A signed trail and evidence packs ready when someone needs to check what happened.

Where the controls apply

People and roles
Who can see, change and approve records in each part of Figena.
AI requests
What may be sent to a model, and what is held back.
Agents and tools
What an agent or outside tool may read and write, and when it needs approval.
Audit and evidence
The signed record of what happened, and the packs that show it.

How an AI action is governed

  1. Someone asks Cilila, or an agent, to make a change.
  2. Guard checks the role and scope, and classifies any data bound for a model.
  3. The change is prepared as a draft, not applied.
  4. A person with the right role reviews and approves it.
  5. The change is made, and every step is written to the signed audit trail.

Research behind the platform

All research