How it works

You do not read BONSAI from cover to cover. You answer nine questions, and those answers activate the controls that apply to you.

Role bounds the solution space; risk and regulation select within it; size changes the implementation, not the obligation.

The rule, without formulas

1. First, what concerns you

   What you do with each system (use it, integrate it, or train it) and what that system does (decide about people, act on its own, process personal data) decide which controls concern you at all. A control that does not concern you is not shown as an obligation: it stays dormant, with its reason in plain view.

2. Then, how far

   How serious a failure would be, whether it can be undone, how many people it reaches, whether the use is high-impact, and the rules that bind you set a letter: B, E or A. That letter says how deep you go into each area: with B, the Baseline controls; with E, the Enhanced ones too; with A, all of them.

3. Finally, in what form

   The size of the organisation neither removes nor adds controls. It decides the implementation mode: light, standard or full. The Baseline is never waived.

The two letters

Each system has its own letter, and the organisation has one too. Controls met once for the whole organisation follow the organisation letter; the rest follow the system letter. That is why a small organisation with one high-impact use has to strengthen that system without building the governance machinery of a multinational.

The nine questions

About the organisation, answered once

A1 Which rules bind you in what you do with AI?

If you operate in more than one country, think of the most demanding one.

• Only general data-protection law applies to our use of AI → Letter B
• Sector regulation applies to us, or rules for organisations using AI in regulated decisions (employment, credit, access to services) → Letter E
• We provide AI systems or models to others under rules for high-risk or general-purpose AI → Letter A
• I do not know

A2 How large is the organisation, and which dedicated functions does it have?

Dedicated functions: security, legal, privacy, compliance.

• Micro or small, with no dedicated functions → light
• Medium, with some dedicated functions → standard
• Large, with dedicated functions → full

A3 Which controls do you already have in place?

• This is not a question in the questionnaire: your own checklist answers it. What is already in place orders your plan, but never removes a control.

About each AI system

B1 What do you do with this system?

• We use it exactly as delivered: an assistant, a SaaS tool, an AI feature inside software we bought → consumer
• We integrate or configure someone else’s model into our processes: an API, connectors, retrieval over our documents, agents → deployer
• We train, fine-tune or substantially adapt it → developer

B2 Does it score, rank, select or decide about people?

• Yes → person-affecting

B3 Does it act on its own, rather than only answer?

• Yes → agentic

B4 Does it process personal data?

• Yes → personal-data

B5 If it goes wrong, what are the consequences?

• Severity of the harm: Minor: an inconvenience or an error corrected in the normal course of work / Significant: a financial loss, unfair treatment or reputational damage / Severe: harm to someone’s health, rights or safety
• Can it be undone?: Easily: the effect is reversed within the same process / With difficulty: it needs separate intervention and leaves a trace / Irreversible: the effect cannot be reversed
• People affected: Few: a handful of internal users or customers / Many: a whole team, a customer segment or a pool of applicants / A very large number: the whole workforce, the whole customer base or the public
• The worst of the three answers counts.

B6 Is it used in a high-impact setting?

• Yes → Letter A

First stage and target

Your first stage is every Baseline control your answers activate: where you begin. Your target is the complete set that applies to you. The first stage is not optional and the Baseline is never waived.

Dormant controls

Controls that do not apply to you today are not hidden. They are shown with their reason (your role, a flag, or your letter), so you can see what you would take on if your situation changes.

When to answer again

Once a year, and in addition whenever any of these changes:

• a change of role: you begin to integrate or to train
• the provider changes the model
• a new use, or a use whose criticality changes
• a new rule or a new jurisdiction
• you discover AI that was not registered
• a significant incident

Go to the assessment The framework