An AI-generated application has no legal regime of its own. The texts that apply depend on three things: the data it processes, what the company does with AI (use it, provide it, embed it) and the sector of the company. This page is for decision-makers, business managers and beginners, with no legal or technical prerequisite. It gives the questions to ask, then goes through the GDPR, the AI Act, NIS2, the CRA and DORA. It gives no legal advice: the qualification of a case belongs to the legal department. For personal data, the data protection officer (DPO) advises, and the company, as controller, decides.
What you will learn
- Ask the scope questions before citing a regulatory text.
- Recognise what the GDPR triggers as soon as an application processes personal data.
- Distinguish a coding assistant user from a provider under the AI Act.
- Know where to read what NIS2, the CRA and DORA say, without confusing them.
- Explain why no scanner establishes compliance.
Which questions should you ask before citing a text?
You do not say "we are subject to this text" before answering five questions. An acronym heard in a meeting proves nothing: each text has a precise scope, date and audience. The five questions fit on one sheet.
- Which legal entity is concerned, in which country, in which sector?
- What data does the application process: personal, sensitive, only internal?
- Who uses it: the author alone, a team, the whole company, outside people?
- What role does the company hold: does it use an AI tool, supply a product to customers, embed a model in its application?
- Which version of the text is in force, on what date?
These questions reuse the criteria of the lesson Classify an application (data, flows, users, criticality) and add the legal role. They also decide who answers: legal for reading the texts, the CISO (the person in charge of information security) for security. For personal data, the DPO advises; the company, as controller, decides. The lesson Governance details this split.
What does the GDPR trigger when an application processes personal data?
The GDPR applies as soon as the application processes personal data, whether the code comes from an AI or from a developer. The General Data Protection Regulation, regulation (EU) 2016/679, has applied since 25 May 2018. Personal data is any information that makes it possible to identify a person: a name, an email address, a phone number. The controller is the person or entity, here the company, that decides why and how this data is processed. In the running example of this course, the "purchasing portal" reads supplier contacts: this is personal data.
The table links each step to its article and to what triggers it.
| Step | Article | What triggers it |
|---|---|---|
| A legal basis (a ground provided by law: consent, contract, legitimate interest...) | Article 6 | Any processing of personal data |
| A contract with the supplier that processes the data on behalf of the company (the processor) | Article 28 | An external service, such as an AI assistant, receives this data |
| A framework for transfers outside the European Union | Articles 44 to 50 | The service sends or stores the data outside the Union |
| An impact assessment (DPIA) before the processing | Article 35 | Processing likely to result in a high risk for individuals |
| Notification of a breach to the supervisory authority, the CNIL in France | Article 33 | Personal data breach likely to result in a risk for individuals |
The table reads line by line: a single application can trigger several lines, and each line has its own owner. The full text of Articles 28, 33 and 35 is in chapter IV of the GDPR reproduced by the CNIL (in French).
Two points are often confused. The first: sending personal data in a prompt to an external service is processing. If the service acts for the company, Article 28 requires a contract. A free assistant used without a contract does not meet this condition. The second: the DPIA is not systematic. Article 35 reserves it for processing likely to result in a high risk, and the CNIL publishes lists of processing that require one. Carrying it out, or deciding that it is not necessary, is a decision of the company as controller (Article 35), taken with the advice of the DPO (Article 39). It belongs neither to the developer nor to the tool.
What does Article 33 say in case of a leak?
The controller notifies the breach to the CNIL without undue delay and, where feasible, no later than 72 hours after having become aware of it, unless the breach is unlikely to result in a risk for individuals. This is Article 33 of the GDPR. A delay beyond 72 hours must be justified. The processor, for its part, notifies the controller without undue delay. Communication to the individuals concerned falls under another article, Article 34, with its own threshold.
A breach is not necessarily a theft. A loss of availability of personal data can be one, even without exfiltration. Conversely, not every outage is one: the risk for individuals must be qualified. The 72-hour deadline applies only to a personal data breach: it holds neither for every security incident, nor for the obligations of NIS2 or DORA, which have their own deadlines. NIS2, for example, requires an early warning within 24 hours, then an incident notification within 72 hours. On supplier checks, the guide TPRM for technical teams lists the notification commitments to ask for.
What does the AI Act require from a company that uses a coding assistant?
A company that uses a coding assistant is a deployer under the AI Act, and its main obligation is the AI literacy of its staff (Article 4). The AI Act is regulation (EU) 2024/1689. It distinguishes the provider, which develops an AI system or places it on the market, from the deployer, which uses it under its own authority. The heavy documentation and transparency obligations target the providers of high-risk systems and of general-purpose models, that is, the vendors. They do not fall on a company that merely uses an assistant to write code.
The regulation was amended by regulation (EU) 2026/1744, called the "AI omnibus", in force since 27 July 2026. The table gives the articles that concern a deployer and their scope.
| Article | What it requires | Scope |
|---|---|---|
| Article 4 | Take measures to support the AI literacy of staff, without guaranteeing a specific level to each person | Providers and deployers of AI systems |
| Article 26 | Deployment obligations: use in line with the instructions, human oversight, log retention | Deployers of high-risk AI systems only |
| Article 50, paragraphs 3 and 4 | Inform exposed people or the public in specific cases (emotion recognition, biometric categorisation, deepfakes, texts published to inform the public) | Some deployers, depending on the use case |
According to the omnibus, the rules for the high-risk systems of Annex III apply from 2 December 2027, those of Annex I from 2 August 2028. The page Vibe coding summarises the same framework. It specifies that the AI Act does not require the user company to document its prompts.
The situation changes if the generated application is itself an AI system, for example an assistant that evaluates candidates or scores customers. You then have to ask whether the use falls within the list of high-risk cases and whether the company becomes the provider of that system. This qualification is made with legal, not with a tool.
Where can you read what NIS2, the CRA and DORA say?
NIS2, the CRA and DORA each have their own page on this site: we do not repeat them here, we link them to the five questions at the start. They do not target the same entities or the same objects, which is why the application must be qualified first.
| Text | What it targets | Question that activates it | Page |
|---|---|---|---|
| NIS2 (directive (EU) 2022/2555) | Essential and important entities of certain sectors | Question 1: entity and sector | NIS2 directive (in French) |
| CRA (regulation (EU) 2024/2847) | Products with digital elements made available on the market | Question 4: does the company supply a product? | Cyber Resilience Act (in French) |
| DORA (regulation (EU) 2022/2554) | Financial entities and their ICT service providers | Question 1: financial entity or not | DORA regulation (in French) |
Software written for purely internal use, never supplied to third parties, is in principle not placed on the market within the meaning of the CRA. An application that the company sells or distributes to its customers may be. The CRA speaks of supply in the course of a commercial activity, paid or free: software shared with subsidiaries or contractors is qualified with legal. The reporting obligations of the CRA apply since 11 September 2026, the others from 11 December 2027.
A word on DORA, which causes confusion. The DORA regulation (Digital Operational Resilience Act) is a European law of the financial sector, in application since 17 January 2025. The DevOps DORA metrics are a performance measure of software delivery, with no legal value. The two are unrelated: the page DORA metrics covers the second.
Does a scanner establish compliance?
No scanner establishes compliance with a text. A secret scanner, a code analyser (SAST, which reads the code without running it) or a dependency tool produce technical evidence: a detected secret, a spotted flaw, a list of components. Compliance is a legal qualification: it depends on the scope, the role of the company and the text in force. A green report does not say that a processing contract exists, that a DPIA has been decided or that a notification is ready.
Compliance is demonstrated by linking each established obligation to a piece of evidence, an owner and a review date. The table shows this link for the purchasing portal, with what the course can prove and what is left to legal.
| Candidate obligation | Applicability | Useful evidence | Who qualifies |
|---|---|---|---|
| Legal basis for supplier contacts | To confirm: processing of personal data established | Entry in the record of processing activities | Controller, DPO |
| Contract if an external service receives this data | To qualify depending on the service | Signed processing contract | Legal, DPO |
| DPIA | To decide: high risk or not | Written decision, with the analysis if one is carried out | Controller, with the advice of the DPO |
| Breach notification | If a breach is established | Procedure and usable logs | Controller, CISO, DPO |
| CRA | Only if the application is supplied to third parties | Qualification decision | Legal |
When applicability is not established, you write "to qualify" rather than "not applicable". The absence of evidence is not evidence of the absence of an obligation. An internal waiver never removes a legal obligation: management has no power to rewrite a regulation.
Key takeaways
- No text applies to "an AI-generated application" as such: the scope is established by the data, the role, the sector and the date.
- The GDPR applies as soon as the application processes personal data: legal basis (Article 6), processing contract (Article 28), transfers (Articles 44 to 50), DPIA if high risk (Article 35).
- A personal data breach is notified to the CNIL without undue delay, if possible within 72 hours (Article 33), unless it is unlikely to result in a risk for individuals.
- A coding assistant user mainly falls under Article 4 of the AI Act, not under the documentation obligations of providers.
- NIS2, the CRA and DORA target specific entities or products: their pages say to whom they apply.
- The DORA regulation has nothing to do with the DevOps DORA metrics.
- A scanner produces technical evidence, never compliance. What is not established is written "to qualify".
Knowledge check
These eight questions cover the scope questions, the GDPR, the AI Act, the NIS2, CRA and DORA references, and the limit of scanners.
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: Governing AI-generated applications: set rules without blocking
A new full quiz with random questions
Work again only on the questions you missed