Compliance · EU AI Act · May 2026 · updated 2026-05-22

EU AI Act.

The EU AI Act enters into force in phases across 2026. Below is the summary of the clauses that touch an AI customer-support assistant — what is your responsibility as deployer, what is ours as provider, and how helpcode is designed so you clear the audit without rewriting your flow.

Hosted in the EU (Frankfurt) No training on customer data Public DPA

In three points

.01

01 · Classification

A technical-customer-support AI typically falls under limited-risk for B2B deployment. helpcode ships the "AI system documentation" pack that satisfies art. 13.

.02

02 · Transparency

Every reply declares it was generated by AI. helpcode includes a configurable disclaimer for widget and voice. Conforms to art. 50.

.03

03 · Traceability

Audit log per answer: prompt, retrieval, model, document version, citation. Exportable in CSV or via API. Conforms to art. 12.

Responsibilities — split

	helpcode (provider)	You (deployer)	Reference
Technical documentation	We provide	Archive	Art. 11
End-user transparency	Widget + voice compliant	Communicate policy	Art. 50
Post-market monitoring	Telemetry + monthly report	Review report	Art. 61
Fundamental rights impact assessment (FRIA)	Template provided	Complete for your case	Art. 27

Frequently asked questions

Do you train models on our data?

No. helpcode uses third-party foundation models (OpenAI, Anthropic, Mistral) via API without opt-in to training. Fine-tuning, when activated, is isolated per tenant and never shared. Specified in DPA art. 4.2.

Where are the data physically located?

Hosted in Frankfurt, Germany. Sub-processors are EU-only by default; the public sub-processor list is updated on every change.

What happens if Anthropic / OpenAI have a breach?

Sub-processor breaches are covered by our DPA: we notify you within 24 hours of becoming aware, and the affected tenant can be migrated to an alternate model provider in hours.

Can I choose the model (e.g. EU-hosted Mistral only)?

Yes. Per workspace, choose Mistral (Paris), OpenAI (EU), Anthropic (EU). Defaults to Anthropic on the EU cluster.

How does data erasure work on request?

Self-service erasure for chat-history data; tenant-wide erasure via support within 30 days, as required by GDPR art. 17.

Last reviewed 2026-05-22 · Email legal

EU AI Act.

01 · Classification

02 · Transparency

03 · Traceability

Frequently asked questions

Other compliance topics