We take your data seriously.

There's a fundamental difference between the public ChatGPT/Claude chat and API access. We work exclusively over the API.

• Data sent over the API is not used to train the model. Contractually and technically — Anthropic codifies this in the Enterprise Agreement.
• No provider employee has routine access to the content of your conversations. Access is possible only in narrowly defined cases (abuse, legal obligation) and under strict audit.
• Data is not shared with third parties. Not anonymized, not aggregated.
• API conversations are not implicitly logged on the provider's side. If we enable logging (for audit trail), retention is set to the agreed period — typically 30 or 90 days, per your internal policy.

The architecture follows current enterprise standards. No improvisation.

• In transit: TLS 1.3 (TLS 1.2 as a minimum), forward secrecy enabled. No legacy ciphers.
• At rest: AES‑256. Keys segregated per client; rotation per policy.
• No data is stored locally on unsecured devices. Endpoints inside the firm (laptops, phones) are protected via MDM recommendations and a zero-trust posture.
• EU data centres. If a specific use case requires CZ/EEA data sovereignty, we handle it via local deployments and European providers.

AI is the lawyer's tool — analogous to other software you already use (DMS, billing, productivity suite). Using a tool doesn't break privilege. What matters is how the tool is configured.

• We recommend codifying AI usage in an internal firm policy — who can use it, for what, and how. We prepare a template for this policy as part of implementation.
• Data classification: some document types are handled by AI by default; others require explicit supervisor approval. We design the policy to fit your practice type.
• Audit trail: who, when, and against which document AI was used. Standard for Enterprise clients; available on request for Professional.
• Our recommendations were reviewed with counsel specialized in compliance and professional duties. The full memo is provided to clients during onboarding.

In a larger firm, different teams work on different matters — M&A, litigation, employment law. Each has different clients, different documents, different confidentiality rules.

We configure AI tools to respect that structure precisely. The M&A team doesn't see the litigation team's files — and vice versa. Access to skills, documents and connectors is segmented per your internal policies, not per the vendor's generic defaults.

Result: AI assistance tailored to each team member's role, with full control on the side of firm management.

AI doesn't add a special GDPR problem — it does, however, put existing principles under sharper light. We have ready answers for what GDPR asks of a provider like us.

• Legal basis: typically art. 6(1)(f) GDPR (legitimate interest of the controller, i.e. you). For sensitive data, a specific basis — handled ad hoc.
• DPA (Data Processing Agreement): we sign our own DPA, layered on top of the sub-processor agreement with Anthropic. Documents are provided before implementation begins.
• Data subject rights: access, rectification, erasure, restriction — all implementable. With one-shot API calls, nothing is persisted long term.
• DPIA (Data Protection Impact Assessment): for projects with significant scope, we prepare a DPIA as part of the design phase.

The gap between what your colleague does on the free version over the weekend and a professional deployment is wide. Concretely: