Review one agent action

Agent Action Security Review

Before your agent takes an important action, understand the controls around it.

Engineering teams and AI implementation partners preparing to launch or hand over an agent that can issue refunds, change records or grant access.

One consequential agent or automation action. Permissions, approvals and execution evidence only.

What you receive

An action map showing the trigger, executing identity, permissions and approval steps for the agreed action.
A review of the execution path, permissions and approval boundary.
Prioritised control gaps linked to the evidence chain, with practical recommendations.
Explicit unknowns: missing evidence and questions the review cannot resolve.
A written report and readout to help your team decide what to change.

One consequential agent or automation action only; multi-workflow programmes need a separate quote.
Default operating mode: read, inspect, reconstruct and report.
No platform installation, production modification, destructive testing, exploitation, credential changes, persistence, continuous monitoring or certification that an agent is safe.
Historical synthetic samples are not a full-review sample and do not change the agreed commercial terms.

Non-secret fit check Name the consequential action, systems involved and decision deadline. Do not send credentials, secrets or production data.
Agreed scope and authority Confirm the action, evidence rules, recipients, price and handling in an accepted agreement. The fee is €2,500 excl. VAT.
Manual reconstruction Read, inspect, reconstruct and report against the agreed evidence. No execution of the reviewed action.
Report and readout Delivery timing agreed in the signed statement of work. Your team owns remediation decisions.

Other review