GDPR-Compliant AI: EU Data Residency & PII

Rolling out AI agents across your organization means sending real business data — often including personal data — into third-party model APIs. If you're a DPO or security lead, that single fact is where most AI adoption plans stall.

Why GDPR and AI don't mix by default

Large language models are built to ingest whatever text you send them, and most consumer AI tools were not designed with Article 5's data minimization principle in mind. The moment an employee pastes a customer email, a support ticket, or a contract into a chat window, personal data may leave your control — and possibly the EU — with no record of what was sent, to whom, or why. GDPR doesn't prohibit AI use, but it does require you to know exactly where data goes, who processes it, and under what legal basis. Most teams adopting AI ad hoc simply can't answer those questions, which is why AI governance has become a standing agenda item for DPOs.

PII redaction at the gateway

AgentWorks addresses this at the point of exposure: every request passes through a gateway that detects and redacts personal data before it reaches any third-party model provider — OpenAI, Anthropic, Google, or Mistral. The data is masked, sent for processing, and unmasked in the response where appropriate, so the model never sees the raw PII in the first place. It's worth being direct about the limits here: free-text PII detection reduces exposure meaningfully, but no pattern-based detection is a 100% guarantee against every edge case. The stronger, structural guarantee sits underneath it — EU hosting and zero-retention, no-training contracts with model providers, so even in the rare case something slips through, it isn't stored or used to train anything. You can read more about how the platform handles data in the Trust Center.

EU data residency and hosting

Where your data physically sits matters as much as what's redacted. AgentWorks' core infrastructure — database, authentication, and storage — runs on Supabase in the EU region, and model calls route through EU endpoints wherever the provider offers them. For organizations with stricter residency requirements, Enterprise plans support self-hosting or a private cloud deployment, keeping data entirely within your own infrastructure rather than a shared multi-tenant environment. This gives DPOs a clear answer to the "where does it live" question that's usually impossible to get from general-purpose AI tools.

No training, zero retention with model providers

A recurring concern with AI vendors is whether your prompts end up shaping someone else's model. AgentWorks does not train models on your data, and contracts with underlying model providers are zero-retention — inputs are processed and discarded, not stored for future training runs. Separately, day-to-day chat messages inside AgentWorks are not auto-deleted (so your team retains conversational history and context), while the compliance audit trail is retained for approximately seven years to support regulatory and audit needs. These are two different retention clocks serving two different purposes, and it's worth understanding both before you set internal data retention policy.

Audit trail and data subject rights

Every interaction — inputs, the model used, tool calls invoked, approvals granted, and outputs returned — is written to an immutable, append-only audit log, exportable as CSV or JSON. That log is what turns "we think our AI usage is compliant" into something you can actually demonstrate to a regulator or an internal auditor. Data subject rights — access, rectification, and erasure — are handled through the dashboard on a per-request basis, so a DPO can action a GDPR request without needing an engineering ticket. Explore the Trust Center for the full detail on logging and export formats.

DPA and sub-processors

A Data Processing Agreement under Article 28 is available on request, and AgentWorks publishes its sub-processor list openly: Supabase, Vercel, Railway, the model providers, Stripe, and Resend, each covered by its own DPA. Infrastructure sub-processors run in EU regions; Stripe processes payments across EU and US as is standard for payment infrastructure. AgentWorks does not hold ISO 27001 or SOC 2 certification today — both are under evaluation, not claimed — and the platform describes itself as EU AI Act-ready rather than certified compliant, which is a distinct and separate framework from GDPR. If you're evaluating AI vendors against both regulations, the EU AI Act guide covers where that framework overlaps with, and diverges from, GDPR obligations.

Summary: GDPR-compliant AI adoption comes down to three things you can verify: PII is redacted before it reaches third-party models, data resides in the EU with no-training contracts, and every interaction is logged in an exportable audit trail with a DPA backing it. AgentWorks is built around all three, honestly scoped — no invented certifications, no promises the architecture can't back up. See transparent pricing or browse AI agent templates to see it in practice.

Frequently asked questions

Do you train on my data?

No. AgentWorks does not train models on your data, and model provider contracts are zero-retention, meaning inputs are processed and discarded rather than stored for training.

Where is my data stored?

Core infrastructure — database, auth, and storage — runs on Supabase in the EU region, and model calls use EU endpoints where providers offer them. Enterprise customers can self-host or use a private cloud to keep data fully within their own infrastructure.

Is AgentWorks ISO 27001 or SOC 2 certified?

No, not currently. Both are under evaluation but not held as certifications today. AgentWorks is built GDPR-by-design and is EU AI Act-ready, and a Data Processing Agreement is available on request — but we won't claim certifications we don't hold.