Classifying an application means answering four simple questions: what data it handles, what it does in the information system (the IS, the set of IT tools of the company), who uses it, and what happens if it stops. The most severe answer gives a class from C1 to C4, and the class decides the controls, the review, the approvals and the length of the authorisation. This page is for decision-makers, business managers and beginners: the lesson and its quiz need no technical prerequisite, while the optional lab needs a terminal and Python. The page includes a classification exercise to do with a pencil.
What you will learn
- Explain why an application is classified before it is hosted.
- Score an application on the four criteria of the grid.
- Calculate its class, from C1 to C4, with the highest-criterion rule.
- Read what each class requires.
- Explain why a declared class that is too low is rejected.
- Recognise the cases where an application must be reclassified.
Why classify an application before hosting it?
You classify to give each application the level of control it deserves, no more and no less. A prototype that its author alone opens needs no barrier. An application that writes into the accounting system needs many. Without classification, everything gets the same regime: too heavy for small uses, which end up bypassing the platform, and too light for large ones.
The class is the entry point of the admission path, that is, the sequence of controls and approvals an application must pass to reach the hosting platform described in Host and isolate applications built with AI. It also extends the distinction between experiment and operated application from the Shadow AI page, by making it measurable.
What are the four criteria of the grid?
Four criteria are enough to place an application: the data, the flows to the information system, the users and the business criticality. Each one gets a score from 0 to 3. The table gives the possible values.
| Criterion | Score 0 | Score 1 | Score 2 | Score 3 |
|---|---|---|---|---|
| Data | none | internal | personal | sensitive |
| Flows to the IS | none | (not applicable) | read | write |
| Users | the author alone | one team | the whole company | outside people |
| Business criticality | none | useful | important | critical |
The words of the table have a precise meaning. Internal data are working documents with no identifiable person. Personal data are names, email addresses, contact details of supplier contacts. Sensitive data cover, in the sense of this grid, health, payroll, secrets and payment data. This is not the definition of the "special categories of personal data" of Article 9 of the GDPR. That article covers racial or ethnic origin, political opinions, religious or philosophical beliefs, trade union membership, genetic data, biometric data that uniquely identify a person, health, sex life and sexual orientation. Payroll, payment data and technical secrets are not on that list.
A read flow means the application reads the ERP through an API, an interface that lets one program query another; a write flow means it creates or modifies data of the information system.
For criticality, a useful application is one whose stoppage is only a nuisance. An important application is one whose stoppage hinders a department for several days. It is critical when its stoppage blocks an activity of the company the same day. The table reads in columns: the higher the score, the greater the possible damage.
How do you calculate an application's class?
The class comes from the highest criterion: a single criterion at 3 is enough to put the application in C4. No average is taken, because an average would hide the weak point that matters. Score 0 gives class C1, score 1 gives class C2, score 2 gives class C3 and score 3 gives class C4.
Take the purchasing portal (named portail-achats in the lab), the invented
case of this course: an application built by a buyer to read the orders of the
ERP, the company's management software. It handles personal data of suppliers
(score 2), reads the ERP (score 2), is used by the purchasing team (score 1)
and its stoppage is important (score 2). The highest criterion is 2, so the
application is C3. If it wrote into the ERP, the flow score would become 3
and the class C4, with nothing else changing.
Two rules complete the calculation. When in doubt about a criterion, you keep the higher score. A class can always be raised; it is lowered only by a written decision of the CISO, who decides the class. A level of control is never lowered by oral agreement or just to save time, and you never go below the class the grid calculates: only a doubt resolved on a criterion can bring a class raised out of caution back to the calculated class.
What does each class change?
The higher the class, the more the controls, the review, the human approval and the renewal frequency increase. The table summarises what the course retains. These thresholds are a starting point for our exercises: a real company adjusts its own to its context.
| Class | Required controls | Review and approval | Maximum authorisation period |
|---|---|---|---|
| C1, personal experiment | No platform control, since the platform does not host the application; the rule of fictitious data still holds | None | Not applicable |
| C2, team use, internal data | Basic controls | A peer review before merge, recorded validation by the technical owner | 12 months |
| C3, company application | Controls of C2, artefact signature, deployment by digest, flows to the IS only through the gateway, logs | Human review of the diff and of the file, one decision signed by an authorised risk owner other than the requester | 6 months |
| C4, critical, external or writing into the IS | Controls of C3, testing of the running application, monthly requalification, withdrawal plan | Review by a person who did not write the request, two signed decisions: that of the manager of the requesting department, other than the requester, and that of the CISO or their delegate, who signs as the authorised risk owner | 3 months |
The basic controls of C2 are secret detection (Gitleaks), static code analysis (SAST, with Semgrep for example), dependency analysis, pinned versions and the inventory of components (SBOM). The test of the running application required in C4 is DAST. A deployment by digest designates the artefact by its digest, a fingerprint of its content that cannot be changed, rather than by a tag that can be moved.
The diff review is the reading of the lines of code added or modified, and the merge is the moment when they enter the version of the application. An authorised risk owner is a person to whom management has delegated, in writing, the power to accept a risk: the lesson Governance details the roles.
The period is not a detail. It forces a regular review and deals with the departure of the author described in Risks of AI-generated applications: an authorisation that expires forces a new owner to be named. Ownership follows the same progression, with a business owner and a technical owner named from C2, both authorised in C4. If a data protection impact assessment is mandatory, it falls to the controller, with the advice of the data protection officer, outside this table. The lesson GDPR, AI Act, NIS2, DORA and CRA details this case.
Why does the platform reject a declared class that is too low?
The request declares a class, the platform calculates it, and the higher of the two wins. The class declared in the application sheet is a request, never a decision. It expresses what the requester believes, and the requester has an obvious interest in seeing the application classified low, since low levels demand less effort.
The platform therefore recalculates the class from the answers given on the four criteria. If the declared class is lower than the calculated class, it rejects the request. It does not correct silently: the rejection forces the requester to reread their answers and to understand what raised the class. The purchasing portal declared as C2 is rejected, because the calculation gives C3.
This rule accuses nobody of bad faith. A buyer who does not tell internal data from personal data sincerely underestimates their application. The automatic calculation catches the mistake, and the grid gives the requester the words to understand it.
How do you practise classifying?
You practise by scoring four requests yourself before reading the answer. Here are four requests invented for the exercise. For each, score the four criteria, then look for the highest one.
- An analyst builds on her own machine a dashboard that only she opens, fed by a file of fictitious data.
- A support team builds a tool that summarises internal meeting minutes, with no customer name, useful to the team but harmless if it stops.
- The purchasing portal described above, which reads the ERP without writing into it.
- A human resources department builds an expense-claim form, open to the whole company, that creates entries in the accounting system from employee data. Its stoppage hinders the department for several days.
The table gives the scores and the expected class. Check the fourth row above all: a single criterion at 3, the write flow, is enough to make it a C4.
| Request | Data | IS flow | Users | Criticality | Class |
|---|---|---|---|---|---|
| 1. Personal dashboard | 0 | 0 | 0 | 0 | C1 |
| 2. Minutes summary | 1 | 0 | 1 | 1 | C2 |
| 3. Purchasing portal | 2 | 2 | 1 | 2 | C3 |
| 4. Expense claims | 2 | 3 | 2 | 2 | C4 |
If one of your answers differs, find the criterion that made you hesitate. It is almost always the distinction between internal data and personal data, or between read and write.
When should an application be reclassified?
An application is reclassified every time one of the four criteria changes. The class describes the application at the moment it was scored, not forever. The purchasing portal classified C3 becomes C4 the day someone gives it the right to write into the ERP, even for a minor function.
Three changes come up often. Opening the application to new users changes the users criterion. Adding a function that modifies data changes the flow. The arrival of more sensitive data, for example the addition of a bank details field, changes the data criterion. Each of these cases restarts the calculation and, if the class goes up, all the controls and approvals of the higher class.
Put it into practice
This lab is optional and needs a terminal, Python 3.11 or later, and the ability to fill in a TOML file and run a Python script.
The lab takes four requests, including that of the purchasing portal, each arriving with the class its requester asks for. You classify them with the grid, you reject those whose declared class is too low, then you apply to each application the policy of its class. The validation compares the class calculated by a script with the declared class, and checks that each application carries the policy that matches its risk.
Key takeaways
- Classification gives each application the level of control it deserves.
- Four criteria, scored from 0 to 3: data, flows to the IS, users, business criticality.
- The highest criterion gives the class, from C1 for score 0 to C4 for score 3.
- When in doubt about a criterion, you keep the higher score; a class is lowered only by a written decision of the CISO, never below the calculated class.
- The higher the class, the more controls, review, signatures and renewals you add, from 12 down to 3 months.
- The declared class is a request, the platform calculates it, and the higher of the two wins.
- An application is reclassified as soon as a criterion changes.
Knowledge check
These eight questions cover the four criteria, the calculation of the class and what each class requires.
Knowledge check
Check what you have learned with this interactive quiz
How it works
- The timer starts when you click Start
- Multiple choice, true/false and short answer questions
- You can move between questions
- Detailed results are shown at the end
Starts the quiz and the timer
Review
(0/0)Score by domain
A domain groups several guides and gets several questions, so its score is more reliable than a per-topic score, which often rests on a single question.
Summary
Strengths
No topic passed at 80% on at least two questions yet.
Weak areas
No topic below the pass mark.
Recommended next
Hints before you try again?
Continue the course: GDPR, AI Act, NIS2, DORA and CRA: what applies?
A new full quiz with random questions
Work again only on the questions you missed