> ## Documentation Index
> Fetch the complete documentation index at: https://docs.odella.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Security review checklist

> Collect evidence for worker access, approvals, credentials, browser use, and offboarding before production use.

Use this checklist with your security owner and the accountable manager of each production AI Employee. A configured control is not the same as a tested control: record what you inspected, the identity used, the expected result, and what happened.

## Review the security boundaries

| Review area | Customer checks | Evidence to keep |
| - | - | - |
| **People and roles** | Active organization, ordinary member roles, team scope, broad administrator access. | Role capability selections and dated membership review. |
| **Worker audience** | Primary team, required role, manager exceptions, and any other applicable grants. | Saved Governance configuration plus allowed/denied member tests, including a direct-link test. |
| **Editing** | Only the necessary people can change instructions, tools, channels, approvals, and recurring work. | Editor-area review and denied-edit tests. |
| **Connected actions** | Exact operations, policy preset, saved allows/blocks, and approval outcomes. | Test records, operation names, and current requests/results—not only a broad preset label. |
| **Outbound messaging** | Current policy, saved recipient rules, expected channels, and attachments. | Safe send/block tests and reviewed exceptions. |
| **Credentials** | User versus account connection ownership, explicit Vault grants, and provider-side permissions. | Connection inventory and scope review; never the secret values. |
| **Unattended work** | Accountable manager, current authorization, schedules, event actions, and escalation paths. | Low-risk task test and manager-handover review. |
| **Signed-in browsing** | User-scoped website sessions, provider permissions, output destinations, and sign-out behavior. | Separate browser tests; connected-tool results do not prove browser restrictions. |
| **Conversation privacy** | Worker conversation sharing and organization administrator read-only viewing policy. | Deliberate choices, privacy review, and representative tests. |
| **Offboarding** | Independent grants, manager reassignment, Vault grants, integrations, and provider session/key revocation. | Removal checklist and retests after access changes. |

Use [Create a team-restricted AI Employee](https://docs.odella.ai/platform/secure-employee-tutorial.md), [Read-only Finance worker](https://docs.odella.ai/platform/finance-read-only-tutorial.md), and [Manager handover](https://docs.odella.ai/platform/manager-handover-tutorial.md) as worked examples.

## Know what Activity can establish

The [Activity Log](https://docs.odella.ai/workforce/activity.md) helps authorized users review worker conversations, tasks, workflows, and service events. Use it alongside task details, approval requests, and your provider's own records to investigate what happened.

Do not treat the worker Activity Log as proof that:

* Every denied access attempt, browser action, or network request is recorded there.
* All records have a particular retention period, immutable storage, or a security-export format.
* A missing activity item means an external action could not have occurred.
* A completed task alone proves that every provider-side change succeeded exactly once.

Activity access and export are separate capabilities. Grant them only where needed, and confirm available export controls and record coverage before including an export in an audit process.

## Request deployment-specific evidence

For controls your team cannot verify from the product UI, coordinate a review through the [Trust Center](https://www.odella.ai/legal/trust) or [support@odella.ai](mailto:support@odella.ai).

Ask for the applicable, current evidence for:

* Worker and browser isolation, network access, and any private-connectivity requirements.
* Data residency, international processing, and customer-selected providers.
* Platform security logging, available customer audit evidence, retention, and incident support.
* Backup, recovery, deletion, and assurance commitments that apply to your agreement.
* The exact supported scope of approval and revocation controls for your use case.

Product source, screenshots, and architectural descriptions explain design; they do not replace deployed verification or contractual commitments. Do not infer universal zero-egress execution, comprehensive data-loss prevention, or per-click browser approval from a general security description.

## Maintain a review record

For each important control, record:

1. **Claim:** What restriction or behavior your organization needs.
2. **Scope:** Worker, team, connection, provider identity, and operation covered.
3. **Owner:** Who maintains the configuration and approves changes.
4. **Evidence:** Configuration reviewed, test performed, result, and date.
5. **Exception:** Any unresolved limitation and the decision to avoid, mitigate, or accept it.
6. **Recheck trigger:** Role changes, new tools, provider-scope changes, manager transfers, and product changes.

Keep secrets, private approval links, and unnecessary customer data out of this record. Review it on your organization's schedule and after consequential changes.

For legal and assurance resources, see [Security and data protection](https://docs.odella.ai/platform/security.md).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.