GDPR for startups starts with the first person in the EU whose personal data your product processes, not with a company-size threshold or a funding round. The rules apply to a two-person team running a beta, and to a startup outside the EU that offers its product to people in the EU.

This guide is for founders whose product was built with AI coding tools. It covers five GDPR gaps to check in an AI-built product, the minimum set of documents to have before launch, three checks you can run without reading code, and when to bring in a privacy lawyer, a data protection officer or a technical review. The question that matters is not only whether your privacy policy is correct, but whether your product does what that policy says. That question is part of shipping AI-built products safely. This article is general information, not legal advice.

TL;DR

A startup meets its GDPR duties when its documents describe what the product actually does with personal data, and the product can act on those documents: collect only what it needs, protect it, delete it on request and support a breach report. For an AI-built product, check the code and the connected services before you rely on the privacy policy.

  • Five gaps to check: personal data without a purpose, services without a data processing agreement, data readable by the wrong user, data stored outside the EU without a transfer basis, and user requests the product cannot carry out.
  • The minimum set before launch: a privacy notice, a lawful basis for each purpose, a record of processing, a breach procedure that can meet the 72-hour notification window, and a recorded DPIA decision.
  • Three founder checks need no code: map the personal data the product stores, list every service that receives it, and test one deletion request across the database, email tool, analytics and logs.
  • Who to involve: a privacy lawyer gets the documents right, a data protection officer is required only in the cases Article 37 lists, and a technical review checks whether the product matches the documents.

When does GDPR apply to a startup?

GDPR applies to a startup as soon as it processes personal data of people in the EU, and company size does not change that. The General Data Protection Regulation covers organisations established in the EU and, under Article 3(2), organisations outside the EU that offer goods or services to people in the EU or monitor their behaviour there.

Personal data is any information about an identifiable person (Article 4(1)). It covers a name and email address, and also an IP address, a device identifier or a support message that names a customer.

The only size-related relief is narrow. Under Article 30(5), organisations with fewer than 250 employees do not have to keep a record of processing activities, unless the processing is likely to result in a risk to people’s rights and freedoms, is not occasional, or includes special-category or criminal-offence data. A live product that processes user accounts every day is not processing occasionally, so the relief does not cover that processing. EU proposals to simplify parts of the GDPR were still being negotiated in November 2026; check whether any change has been adopted before relying on a size threshold.

In the Netherlands, the GDPR is known as the AVG, and the supervisory authority is the Autoriteit Persoonsgegevens (AP).

Does building with AI tools change your GDPR obligations?

Building a product with AI coding tools does not change your GDPR obligations. Your startup remains the controller: the organisation that decides why and how personal data is processed (Article 4(7)). What changes is where you need to look for gaps.

An AI coding tool turns a prompt into working features: sign-up forms, database tables, analytics events, connections to email or AI model services. Each of those features can process personal data. Whether each field has a purpose, where the data is stored and how it is removed are decisions your team still has to make and check.

That is how the privacy notice and the product can drift apart: the notice records what the founders intended, while the code and its configuration show what the product does. A customer questionnaire, an investor’s due diligence or a user’s deletion request can be the first moment the difference becomes visible.

5 GDPR gaps to check in an AI-built product

5 GDPR gaps to check in an AI-built product

The five gaps below are grouped by what goes wrong under GDPR, not by tool or platform. Any of them can occur in an AI-built product; which ones apply to yours depends on how it was built and configured, so treat the list as places to look, not as findings.

Personal data collected without a defined purpose

GDPR requires each piece of personal data to have a specified purpose and to be limited to what that purpose needs (Article 5(1)(b) and (c)). In an AI-built product, generated sign-up forms, database tables and analytics events can include fields that no feature uses, and nothing removes those fields later.

A prompt for an onboarding flow can return a form that asks for a phone number, a date of birth and a full address when the product only needs an email address. Logging added for debugging can store complete request bodies, including whatever users typed into a form. Article 5(1)(e) adds storage limitation, so each field also needs a retention period. The signal to look for: database fields that no screen or feature reads, and no retention period written down anywhere.

Personal data shared with services that have no processor agreement

A service that processes personal data on your startup’s behalf is a processor, and GDPR Article 28 requires a contract that sets out what it may do with that data. An AI-built product can connect several such services, such as hosting, authentication, email, analytics, error logging and AI model APIs, without anyone checking the contract terms for each.

The Dutch supervisor is direct about it: you have to conclude a processing agreement with your processor. That contract is the data processing agreement (DPA). Some services include a DPA in their terms; others need one signed separately.

An AI model API needs the same check when prompts contain personal data, for example when a support ticket is sent to a language model for a summary. The provider’s role depends on how and why it processes the data; where it processes personal data on your instructions, Article 28 applies. Its terms should also say whether it keeps prompts and whether it uses them to train models. The signal to look for: no list of the services that receive personal data, or a list without a DPA status next to each name.

Personal data readable by the wrong user

GDPR Article 32 requires security appropriate to the risk, and access is the first thing to test: a signed-in user must not be able to read another user’s records. In an AI-built product, that protection depends on database access rules and server-side checks, which can be missing or written too broadly.

In a product built on Supabase, for example, row-level security decides which rows each signed-in user can read. If it is switched off, or a rule allows every signed-in user to read every row, one account can retrieve another account’s data. Our guide to Supabase row-level security explains how to check it.

When a flaw like this exposes personal data, it is a personal data breach, and the notification rules in Article 33 apply. Ask your developer: “Can an ordinary account retrieve another account’s records, and which test shows that it cannot?”

Personal data stored outside the EU without a transfer basis

GDPR Chapter V allows personal data to be transferred outside the European Economic Area only on a legal basis, such as an adequacy decision or standard contractual clauses. An AI-built product can move data abroad without a deliberate decision when a hosting, database or AI service was set up in a non-EU region and nobody recorded that choice.

For US providers, the European Commission’s adequacy decision of 10 July 2023 for the EU–US Data Privacy Framework allows transfers to US organisations certified under that framework. The EU General Court upheld the decision, and an appeal is pending before the Court of Justice, so check its status before relying on it.

A transfer with a valid basis still belongs in your records and privacy notice. GDPR does not require EU hosting, but choosing an EU region, where a provider offers one, removes the transfer question for that service. The signal to look for: a service list with no storage region next to each name.

User rights the product cannot carry out

GDPR gives people the right to access, correct, delete and receive a copy of their personal data (Articles 15, 16, 17 and 20). Article 12(3) sets a response time of one month, which can be extended by two further months for complex or numerous requests. An AI-built product without an admin function may only handle these requests through a developer editing the database by hand.

Deletion needs the widest reach, because personal data can sit in several places: the account record, uploaded files, the email tool’s contact list, analytics profiles, error logs and backups. The signal to look for: nobody on the team can say how one user’s data would be deleted everywhere.

Not sure which of these five gaps apply to your product?

An AI-built product review examines security and data in the code, including permissions, cross-user access, database rules and sensitive data flows. Findings come in plain English, with what to fix now and what can wait, so your developer and your privacy lawyer can act on them.

See how an AI-built product review works →

The minimum GDPR set a startup needs before launch

GDPR compliance for startups rests on five documents and decisions that describe the processing and name who acts when something goes wrong. Each one is only as accurate as the team’s knowledge of what the product does.

A privacy notice that matches what the product does

Articles 13 and 14 require you to tell people what personal data you collect, why, on which legal basis, who receives it, whether it leaves the EU and how long you keep it. A template notice that names services your product does not use, or leaves out services it does use, does not meet that requirement. Write the notice from your list of stored personal data and connected services, and update it when a feature adds a field or a service.

A lawful basis for each purpose

Article 6 lists six lawful bases for processing personal data, and consent is only one of them. Processing needed to deliver the service a user signed up for can rest on the contract basis in Article 6(1)(b).

Choosing consent for that core processing creates a problem: people can withdraw consent at any time (Article 7(3)), and a core feature cannot depend on a basis the user can switch off. Keep consent for processing a user can refuse without losing the service, such as a marketing newsletter. Cookie consent on your website follows its own rules, which our GDPR website checklist covers.

A record of what personal data you process and why

Article 30 requires a record of processing activities: the categories of personal data, the purposes, the recipients, any transfers outside the EU and, where possible, the retention periods. For an early-stage startup this can be a single spreadsheet, and the under-250 relief in Article 30(5) does not cover processing that is not occasional. A business customer’s vendor questionnaire or an investor’s due diligence can ask to see the record.

A breach procedure with a named owner

Article 33 requires a controller to notify the supervisory authority of a personal data breach without undue delay and, where feasible, within 72 hours of becoming aware of it, unless the breach is unlikely to result in a risk to people. In the Netherlands, that authority is the Autoriteit Persoonsgegevens.

The AP’s data breach guidance also requires a register of every breach, including breaches you decide not to report. When a breach is likely to result in a high risk to people, Article 34 adds a duty to inform the affected users. A one-page procedure that names who assesses a breach, who decides on notification and where the register is kept keeps the 72 hours for the assessment itself.

A DPIA decision for higher-risk processing

Article 35 requires a data protection impact assessment (DPIA) before processing that is likely to result in a high risk to people’s rights and freedoms. Not every startup needs one, but every startup should record why it does or does not, so the decision is visible when a customer or investor asks.

Article 35(3) names cases where a DPIA is always required, such as large-scale processing of health or other special-category data, and systematic, extensive evaluation of people that leads to decisions with significant effects. The Dutch government’s business portal explains when the obligation applies. For an AI-built product, the moment to re-check is a new feature: adding AI scoring, profiling or health data can change the answer.

How to check your AI-built product against GDPR

A founder can run three checks without reading code, with a developer or the AI coding tool that built the product at hand. The checks show whether the product matches the privacy notice; they do not establish that the product complies with GDPR. For the privacy-by-design checks under Article 25, use our production readiness checklist.

Map the personal data your product stores

Ask your developer, or the AI coding tool that built the product, to list every database table and field that holds information about a person, and what each field is used for. A field without a purpose fails the data minimisation principle in Article 5(1)(c) and still has to be protected and deleted.

Put the answer in a table with three columns: field, purpose and retention period. That table becomes the core of your Article 30 record. Ask: “Which fields does no screen or feature read?”

List every service that receives personal data

Collect the project’s API keys and environment variables, then add every tool billed to the company card. Each service on that list that receives personal data needs a DPA, a recorded storage region and a line in your privacy notice.

Environment variables hold the keys a product uses to connect to outside services, so they show which services the code talks to. For AI model services, ask whether prompts include user content. For mobile apps, third-party SDKs belong on the same list; our guide to GDPR for mobile apps covers them. Ask: “Which services receive personal data, and in which region does each one store it?”

Test a deletion request from start to finish

Create a test account, use the product as a real user would, then delete the account through the product or ask your developer to delete it. Check afterwards whether the data is gone from the database, file storage, email tool, analytics and error logs. Each place where it remains without a reason to keep it, such as a legal retention duty, is an Article 17 gap.

Backups follow their own cycle. Record how long they are kept, and make sure that restoring a backup does not bring deleted accounts back. If the test leaves data in two places, you have two fixes to make before the next real deletion request arrives.

Do you need a DPO, a privacy lawyer or a technical review?

A startup needs a data protection officer (DPO) only in the cases Article 37 lists, a privacy lawyer or consultant to get the documents right, and a technical review when nobody can confirm that the product does what those documents say.

Under Article 37(1), a DPO is mandatory for public bodies and when the core activities consist of regular and systematic monitoring of people on a large scale, or large-scale processing of special-category or criminal-offence data. Outside those cases, name one person who owns privacy decisions.

WhoWhat they answerWhen a startup needs them
Data protection officer (DPO)Ongoing oversight of GDPR compliance; contact point for the supervisory authorityWhen Article 37(1) applies: large-scale monitoring or large-scale special-category data as a core activity
Privacy lawyer or consultantLawful bases, privacy notice, DPAs, DPIA, customer contractsBefore launch, and when entering a new market or adding sensitive data
Technical review, such as an AI-built product reviewWhether the product does what the documents say: what it stores, who can read it, which services receive itBefore launch or the first paying customers, before investor due diligence, and when the product was built quickly with AI tools
Compliance readiness for ISO/IEC 27001, SOC 2 or NIS2Security management evidence for customers and auditorsWhen business customers ask for ISO/IEC 27001 or SOC 2 evidence, or NIS2 applies; not a GDPR certification
Table 1. Who answers which GDPR question for a startup

Start with the three founder checks. If you can answer them, a privacy lawyer can document what the product does. If you cannot, start with an AI-built product review: a lawyer can only describe the processing you tell them about. When business customers also ask for ISO/IEC 27001 or SOC 2 evidence, Sunbytes’ compliance readiness service maps the gaps and builds that evidence; it does not issue certificates.

Does the EU AI Act apply to an AI-built product?

Using AI coding tools to build a product does not, on its own, bring the product under the EU AI Act. The AI Act, Regulation (EU) 2024/1689, regulates AI systems that are placed on the market or used. It becomes relevant when the product itself contains AI features, such as a chatbot or automated scoring of people.

For a startup with AI features, the transparency duties in Article 50 apply from 2 August 2026. Depending on whether you provide or deploy the AI system, they include telling people when they are interacting with an AI system and marking AI-generated content. The Digital Omnibus on AI, Regulation (EU) 2026/1744, gives providers of generative AI systems already on the market four months to adapt their marking, and sets 2 December 2027 for the high-risk rules covering Annex III uses such as recruitment or credit scoring. GDPR keeps applying to any personal data those features process.

How Sunbytes reviews AI-built products for personal-data risks

A privacy notice describes the product you intended to build; the deletion test shows the product you have. When the two differ, the fix sits in the code and its configuration: access rules, logging, connected services and the way data is removed.

Sunbytes reviews AI-built products independently of whoever built them, with engineering and security reviewed together across the code, product behaviour and technical setup. You receive findings with the evidence behind them and a prioritised action plan that separates what to fix now from what can wait. Your team or AI coding tool can act on the technical handoff, or Sunbytes can fix priority issues and check them again.

The review is technical. Sunbytes does not give legal advice or certify GDPR compliance; the findings are written so your privacy lawyer or DPO can use them. Access and information handling run under Sunbytes’ ISO 27001-certified information security management system, with Netherlands-led governance and 15+ years in software delivery across 300+ projects.

Scope depends on your stack, so it is agreed case by case. Talk to us about an AI-built product review before your first paying customers or your next due diligence round.


Author

Thien Le Marketing Executive

Thien Le is a B2B content writer with more than four years of experience in crafting compelling B2B content for a global audience, specializing in how people, technology, and business operations come together. At Sunbytes, she plans, connects insights from HR experts, edits, and uses technology to improve content quality. Her content helps readers cut through complexity and find practical information for hiring, paying, and managing people across borders.

What can we help you move forward?

Tell us what you are working on. We will connect you with the right Sunbytes solution.

[ENG] Submission form (Homepage, Service & Contact us)

This field is for validation purposes and should be left unchanged.
Your Full Name(Required)
Max. file size: 128 MB.
untitled(Required)
Untitled(Required)