# Security & Trust

Support data is some of the most sensitive data a business holds — customer identities, account details, and the substance of their problems. And an AI-native helpdesk raises the bar further: if AI agents can act on tickets, the controls around them have to be real. This article covers the security and governance model for technical evaluators. Operational configuration is in the signed-in documentation.

## Access Control

Staff access is governed by role-based permissions, granular to the capability:

- Separate keys for viewing, creating, editing, and deleting tickets
- Separate keys for reading versus managing the knowledge base
- Separate keys for viewing versus reviewing AI approvals
- A distinct key for managing helpdesk configuration (teams, SLA, portal, AI policy)

That granularity lets you compose roles that fit how your team actually works — a front-line agent, a lead who reviews AI actions, a knowledge manager, an admin — without over-granting.

## Isolation

The helpdesk enforces isolation at two levels:

- **Tenant isolation** — each customer of Karmaflow.ai operates in its own data boundary; one tenant's support data is never reachable from another.
- **Organization isolation** — inside a tenant, one customer *organization* can never see another's tickets. Portal reads are checked on every request and scoped to the requesting user's own company; a ticket a user isn't entitled to see is indistinguishable from one that doesn't exist.

The portal is a separate, token-authenticated surface from your staff workspace, with its own sign-in — customers never touch internal tools, and internal sessions never cross into the portal.

## Provenance and Audit

Every meaningful action is attributable and recorded:

- **Origin traceability** — each ticket and message records where it came from: the chat, call, SMS, email, form, automation, or the specific person that produced it.
- **Attribution** — every message and action is stamped with its author, and explicitly marked as human or AI. The AI never acts anonymously.
- **Audit trail** — ticket mutations and AI actions are written to an audit log, so you can reconstruct exactly what happened, when, and by whom.

## AI Governance

Because AI agents can act, the helpdesk treats AI actions as a governed surface, not a free-for-all. Each proposed AI action is classified — allowed, needs human approval, or forbidden — against a policy you control, with confidence thresholds and sensitive-topic detection that route uncertain or high-stakes actions to a human. Approvals are explicit and logged; a human is always reachable. The full model is covered in **AI-Native Support**.

## Why It Matters

Trust is what lets a support leader turn AI on at all. Karmaflow.ai's answer is the same discipline you'd expect from any serious system of record — least-privilege access, hard isolation, complete provenance and audit — extended to cover AI as a first-class actor. You can give the AI real work to do precisely because the controls around it are real.

> **Need the compliance and configuration specifics?** Role setup, permissions, and the AI governance policy are detailed in the signed-in documentation. [Sign in to Karmaflow.ai](https://omni.karmaflow.ai/auth/login) to read them.