STEP 1 OF 7

Understand the use case

Describe what the agent is for and what it does. Every later step builds on these answers.

1Name the use case and what the agent should achieve
2List each action the agent takes (verb + object) and what it changes
3Answer the context questions; derived lines are prefilled, confirm or change them
4State what the agent must never do

1 · Purpose

Name of the use caseRequiredI01

Short and specific

Example: Supplier invoice payment agent

What should the agent achieve for the business?RequiredI02

Describe the goal, not the technology

Example: Pay approved supplier invoices on time with less manual effort

Where does it sit in the business process? What happens before and after?RequiredI03

Name upstream and downstream steps and systems

Example: After invoice receipt in the ERP; before the bank payment run

2 · What the agent does

List each action separately — one verb and object per action.

3 · Context

I04Derived from your actions:acts and informs

Who is affected by its outcomes?RequiredI05

Include people outside the organization

Example: Suppliers, internal finance team

Organization contextRequiredI06

Industry, size and whether you are a public body or provide public services

Example: Manufacturing, 5,000 employees, private

Where is the agent used? (jurisdictions)RequiredI15

Select all that apply; used only to list legal obligations, never to change the AI-SIL

Example: EU / EEA

Which data does the agent use?RequiredI07

Mark personal, confidential and untrusted external data

Example: Invoices from suppliers (external), vendor master, bank details

Which systems and tools can the agent use?RequiredI08

APIs, applications, code execution

Example: ERP read/write, payment API, email

Does it work with other agents or other organizations' systems?RequiredI09

Name each agent or partner

Example: Receives orders from an ordering agent

Which model or product powers it, and who controls updates?RequiredI12

Vendor-managed or internal

Example: Vendor model, updates by the vendor

Does the agent's output inform decisions about people, money or regulated controls, even if a person makes the final call?RequiredI14

Examples: credit or benefit decisions, hiring, medical or legal advice, compliance checks, fraud or AML reviews

Example: Yes: payment priorities affect suppliers' cash flow

How do humans oversee it today?RequiredI10

Who reviews what, and roughly how many items per hour

Example: AP clerks review exceptions, ~40 per hour

Asked because an action has a human check: how many items can one reviewer handle?

How many decisions does the agent make?RequiredI11

Per day or per month

Example: ~800 invoices per day

4 · Boundaries

What must the agent never do? Which cases are excluded?RecommendedI13

Hard boundaries and known exclusions

Example: Never change bank details; no payments to new suppliers

Facts about your use case

19 of 19

Answer for the whole flow. These decide which grey areas and hazards you see next. "I'm unsure" counts as yes and adds an expert review.

T01

Can the agent execute irreversible actions (payments, contracts, deletions, external messages)?

Usually hard to undo in time

T02

Do decisions affect persons outside the organization (customers, applicants, citizens)?

Runs fundamental-rights screening

T03

Is the deployer a public body, a provider of public services or using AI for credit scoring or life/health-insurance pricing?

Flags a legally mandatory FRIA (EU AI Act Art. 27)

T04

Is the use case possibly in an EU AI Act Annex III area (e.g. employment, credit, education, essential services)?

Flags high-risk obligations; recommends expert review

T05

Does the agent process personal data?

Reuse an existing DPIA where available

T06

Does the agent ingest untrusted external content (web, email, uploaded documents)?

Security-for-safety

T07

Does the agent call tools, APIs or execute code?

Security-for-safety

T08

Does the agent keep memory across sessions or tasks?

T09

Does the flow involve more than one agent?

Flow-level classification; network assessment

T10

Does the flow exchange actions with agents or systems of other organizations?

Cross-firm interface contract

T11

Are third-party models, tools or agent products used?

AI-SIL 3+: recognized certification or equivalent evidence

T12

Do human reviewers handle high volumes (e.g. more than a few decisions per minute)?

Oversight sufficiency test

T13

Does the agent communicate directly with humans (chat, email, voice)?

AI disclosure (EU AI Act Art. 50)

T14

Does the agent optimize a target metric or KPI autonomously?

SOTIF-type risk

Expert review recommended

T15

Can the model or provider change outside the deployer's control?

Re-classification trigger

T16

Can the agent's statements create obligations for the organization or be taken as its official position (prices, refunds, policies, legal, medical or financial information)?

Statements can bind the organization (v0.4)

T17

Can a single input, sensor or data source drive an irreversible or high-impact action without an independent cross-check?

Single point of failure (v0.4)

T18

Do the agent's decisions accumulate exposure over time or across many small actions (positions, inventory, commitments, budget)?

Accumulating exposure (v0.4)

T19

Is the agent a detector, filter, screener or monitor, so that its misses are seen by no one (fraud, intrusion, content or safety screening)?

Omission rule: 'suggests only' gives no E0 for misses (v0.5 D2)

Step 1 complete.

Classification is deterministic — AI only adapts wording and suggests, you confirm.