# The BROCS assessment

> Twenty statements, four per letter. Score the ones that are true today and you get a scoreline.

Source: https://brocs.fyi/resources/assessment/
Framework: BROCS (Build, Run, Observe, Control, Secure), brocs.fyi. CC BY 4.0, attribute to brocs.fyi.

---


Check every statement that is true of your organization **today**. Not planned, not in
someone's roadmap, not true for one team that figured it out on their own. True.

Your score for each letter is the number of statements you could check, so each letter
lands between 0 and 4 and the whole thing reads as a scoreline: `B3 R2 O1 C2 S0`.

Nothing is stored and nothing is sent anywhere. The scoreline lives in the URL, so the
link is the result: send it to whoever needs to see the gap.



## B · Build: Enable people to make things.

- [ ] A new hire can get to a working AI development environment on their first day without filing a ticket.
- [ ] Model API keys are issued, scoped, and rotated by a system, not pasted into a shared document or a personal account.
- [ ] Someone who is not an engineer can build and ship something useful with AI on a supported path.
- [ ] Swapping one AI coding tool for another is a configuration change, not a project.

## R · Run: Give the work somewhere to live.

- [ ] AI apps and agents run somewhere the platform team can name, with the same deployment path as everything else.
- [ ] Databases, object storage, and secrets for AI workloads are provisioned on request rather than hand-built.
- [ ] No production credential is held by an agent that also accepts free-text instructions.
- [ ] You could run the same workload in a different cloud, or on your own hardware, without rewriting it.

## O · Observe: See what it is doing.

- [ ] You can say what an AI application cost last month, broken down by team or by app.
- [ ] You can reconstruct what an agent did in a given run: prompts, tool calls, and data touched.
- [ ] Alerts about an AI app reach the person who built it, not only a central operations queue.
- [ ] You would find out that model output quality got worse from your own measurements rather than from a customer.

## C · Control: Keep a hand on the wheel.

- [ ] Model, prompt, and routing configuration is versioned and reviewed like code, and can be rolled back.
- [ ] Internal AI apps sit behind one front door with SSO, rather than a tunnel or a share link per team.
- [ ] You can move a workload from one model or provider to another centrally, without touching application code.
- [ ] Backups of AI application data exist and a restore has actually been tested.

## S · Secure: Prove it is not going off the rails.

- [ ] Every agent's tool permissions are enumerated somewhere a human reviews.
- [ ] You have a current picture of which unsanctioned AI tools are in use in the organization.
- [ ] AI spend is attributed to a team or cost centre and reported back to them.
- [ ] You could produce audit evidence for an AI system this week without a scramble.

Score one point per true statement. Five letters, 0 to 4 each, read as a scoreline
like B3 R2 O1 C2 S0.