Drel in practice

See a review before you run one.

One review's record, for a fictional AI system: its threat model, control gaps and security clearance. On your system, Drel also checks the controls it can verify in your code and proposes the fixes.

Illustrative example · fictional system

Enterprise procurement agent

Restricted pilot
RAGToolsMCP

An agentic system with write access to procurement workflows: supplier analysis, contract drafting, and purchase order execution. 10 open findings, 10 controls and 5 pieces of evidence still needed.

Inside the sample evidence pack

Three of its pages, as an AI Committee receives them. The system, organization and people are fictional.

Cover of the sample evidence pack: the Restricted pilot decision for the fictional Enterprise Procurement Agent and the conditions it carries
The clearance decision and what it rests on
Findings page of the sample evidence pack: critical findings with their attack path, business impact and evidence state
Findings, with their attack paths and business impact
Controls page of the sample evidence pack: the controls required before the pilot and before production, each with an owner and a due date
Required controls by gate, with owners and due dates

Run a review on your system

Connect the repository and design docs (a written description still works). Drel threat-models the system, checks the controls it can verify in your code, proposes the fixes, and produces the security clearance.

Request access