Security & recovery
These are decisions to make before handing an AI system access to your data or infrastructure.
Write down who can do what.
An agent needs a clear job and limited permissions. Your team needs to know which actions it can take, which need approval, and how to stop it.
Agent permissions and human responsibility+
- Name the human responsible for each system and each approval.
- Start integrations with only the permissions their task needs.
- Keep data restrictions in place when work passes between tools.
- Check the result of an action rather than relying on an agent reporting success.
- Preserve disagreements and failed checks for a person to review.
These are design principles. The implementation needs tests against the access rules agreed for your environment.
Backups are only useful if they restore+
Agree how long an outage can last and how much data can be lost. Then test the restore procedure on the hardware and backup arrangement you will actually use.
| Scenario | What to verify |
|---|---|
| A service fails | How it is detected, restarted and checked. |
| A database needs restoring | Which backup to use, how long restoration takes and whether the data is intact. |
| The machine is unavailable | Where the replacement runs and how configuration, data and access are restored. |
| The usual operator is absent | Whether another authorised person can follow the runbook. |
Recovery targets belong in the scope of your build. No recovery time is promised by this demonstration.
This is a demonstration of the platform. CPLT — the association behind it, and the builds it takes on when a team asks — lives on cplt.tech, along with what a build costs. It starts with a free forty-five-minute scoping call that ends in a one-page written note on what your team can run.