Privacy for AI systems
An AI feature is a new route for personal data: into a training set, into a prompt sent to a provider, into an index a model retrieves from, and out again in an answer. Each route has a privacy question that is easier to ask before launch.
The short version
Privacy for AI systems is the review of where personal data enters an AI system, through training data, prompts, retrieved documents and outputs, and what the organization must be able to say about each.
The security side of the same systems, meaning prompt injection, retrieval isolation and agent permissions, is under AI security; this page is about the personal data in them.
Where personal data enters
Four routes are reviewed: training and fine-tuning data, prompts, retrieved documents, and outputs and decisions.
Training and fine-tuning data needs the same questions as any other use of personal data: where it came from, on what basis it is used, how long it is kept and whether a person’s request can reach it. Technical controls for that data are on fine-tuning and training data security and RAG and vector security.
Automated decisions and the AI Act
Article 22 of the GDPR concerns decisions based solely on automated processing that produce legal or similarly significant effects on a person, and a system that scores, ranks or screens people should be checked against it. The GDPR is on EUR-Lex.
The EU AI Act, Regulation (EU) 2024/1689, is a separate instrument from the GDPR, and an AI system that uses personal data can be subject to both. It is published on EUR-Lex. We do not state its application dates here, because they change; the current text is checked at the start of each engagement. A plain-language article is the EU AI Act for Canadian suppliers.
An impact assessment is the usual tool for a system that profiles or decides about people: see privacy impact assessments.
What you get
- An AI data map. Each AI system, its data routes and the providers that receive personal data, added to the records of processing.
- Provider and transfer review. What leaves your systems to reach a model, on what terms and to which country; see AI use discovery and vendor risk.
- An impact assessment for systems that decide or rank people.
- Governance inputs. Privacy rules for AI use that fit into AI governance readiness.
Who it is for
Organizations adding AI to a product that holds personal data, those whose staff already use AI tools with customer information, and teams preparing to answer a customer’s or regulator’s AI questions. The legal conclusions are your counsel’s; the facts that support them are ours to establish.
Under US law
California’s regulations on automated decisionmaking technology took effect on 1 January 2026, and a business that uses it for a significant decision before 1 January 2027 must comply by that date (CPPA regulations page). Which of your AI features make such decisions is an inventory question; the dates are in California’s 2026 privacy regulations.
How the work is bounded
The scope is agreed in writing before work starts, and the engagement is quoted in writing with it.
SecHB does not issue certifications, attestations or audit opinions: those come from accredited certification bodies, CPA firms and QSAs. The work here is what an organization does to be ready for them.
Nothing here is legal advice. Where a question turns on the law, the work is done alongside the client’s counsel, not instead of them.
Questions we are asked
What does a privacy review of an AI system cover?
Privacy for AI systems is the review of where personal data enters an AI system, through training data, prompts, retrieved documents and outputs, and what the organization must be able to say about each.
Does the GDPR restrict automated decisions?
Article 22 of the GDPR concerns decisions based solely on automated processing that produce legal or similarly significant effects on a person, and a system that scores, ranks or screens people should be checked against it.
Does the EU AI Act replace the GDPR?
The EU AI Act, Regulation (EU) 2024/1689, is a separate instrument from the GDPR, and an AI system that uses personal data can be subject to both.
What about training and fine-tuning data?
Training and fine-tuning data needs the same questions as any other use of personal data: where it came from, on what basis it is used, how long it is kept and whether a person’s request can reach it.
Describe the AI system and the data it sees
Background reading: personal data in AI systems. Say what the system does, what data reaches it and where it is hosted. The reply says what a review would cover. See also all privacy services.