Frequently Asked Questions

No. Data residency (where data physically sits) and data sovereignty (who can legally compel its disclosure) are different things. AWS, Microsoft Azure, and Google Cloud all operate UK data centres — but their parent companies are US corporations subject to the US CLOUD Act 2018. A US court can compel them to produce data from a UK data centre without notifying the UK data subject, often under a gag order. Private AI deploys on infrastructure whose owner sits outside US jurisdiction — a UK-registered provider for UK deployments, a European-owned EEA provider for EEA deployments. Neither has a US parent a US court could compel under the CLOUD Act to produce your data, so the Act does not reach the client data on your server. For the full legal analysis, see our article on the CLOUD Act and UK data sovereignty at /insights/cloud-act-uk-data-sovereignty.

No. We operate at the infrastructure layer only — we hold zero logical access to client databases, prompts, conversations, or model outputs. Our Zero Standing Privileges model means we hold no persistent credentials to your environment. When maintenance is required, temporary just-in-time access is granted by your IT Lead and revoked on completion. All sessions are keystroke-logged and auditable. For a detailed walkthrough of the access model, see /sovereignty/architecture.

It remains on your server, under your control. Your organisation holds the direct contract with the bare-metal infrastructure provider. If you end your retainer with us, your server continues running under your account. We hold no lock-in over your AI workspace, your data, or your infrastructure. The Handover Certificate documents this explicitly — the environment is yours from day one.

Zero-egress AI means no data ever leaves the secure perimeter of your dedicated infrastructure. Unlike public AI services where prompts traverse the open internet and pass through vendor servers, our architecture enforces network-level isolation — there is physically no outbound path for client data. This is achieved through egress-controlled networking on bare-metal GPU nodes.

Public AI services process your data on shared multi-tenant infrastructure controlled by a foreign-jurisdiction vendor. Your prompts — potentially containing privileged client information — are transmitted to, and processed on, servers you do not control, under terms you cannot negotiate. Private AI deploys sovereign, open-weight models on dedicated bare-metal infrastructure within UK/EEA jurisdiction. You hold the direct infrastructure contract. We never see, access, or process your data.

None. In the Default Operating State, Private AI holds no active connection, no credentials, and no means of accessing your systems — enforced at the network level by the Zero Standing Privileges architecture. We configure and deploy your sovereign AI environment, then permanently revoke our own access. We hold zero standing privileges to your running systems, zero access to your data, and zero ability to read your prompts or outputs. This is not merely a policy — it is the architecture.

The DUAA 2025 is UK legislation that imposes new duties on organisations using AI with personal or sensitive data. It strengthens accountability requirements, mandates transparency in automated decision-making, and introduces enhanced enforcement powers for the ICO. Our zero-egress architecture is designed to support compliance with these requirements. See /sovereignty/compliance for a full discussion.

Our standard Sovereign Transition takes 14 working days from contract signature to operational handover. This includes infrastructure provisioning, model deployment, security hardening, integration testing, and team onboarding. The process is designed to minimise disruption to your existing workflows while delivering a fully operational private AI environment.

Our architecture is designed to support compliance with UK GDPR (including Articles 28, 32, and 35), the Data Use and Access Act (DUAA) 2025, Law Society of Scotland practice rules on confidentiality and data handling, and ICAS ethical guidelines for accountancy practices. We provide a Compliance Facilitation Pack including DPIA templates, processing records, and architectural evidence for your regulatory submissions.

All data is processed on bare-metal GPU infrastructure within UK and EEA jurisdictions and stored on encrypted persistent NVMe storage attached to that infrastructure. You hold the direct contract with the infrastructure provider, and you contract us separately — every processor is one you've signed with directly. We engage no sub-processors of our own; the provider discloses its own in the DPA you sign with it. Nothing is hidden behind the scenes. Data residency is set at the infrastructure level you control, not merely by policy or promise.

Zero Standing Privileges (ZSP) means that no person — including Private AI staff — holds persistent access to your deployed environment. After deployment and handover, all administrative access is revoked. Any future maintenance access requires your explicit, time-limited authorisation. This architecturally constrains the insider threat vector and provides a clear audit trail for regulatory review.

We deploy sovereign, open-weight large language models that run entirely within your isolated infrastructure. These are production-grade models that handle the vast majority of professional work, but without the data sovereignty risks. The specific model is selected based on your use case requirements and can be updated or replaced without changing your infrastructure.

We serve any organisation that handles confidential or sensitive data and can't afford to expose it to public AI — law firms, accountancy and financial practices, healthcare providers, housing associations, charities, universities, and others. Our architecture is particularly relevant wherever a duty of confidentiality or a regulator governs how data is handled.

Private AI Ltd acts as a Standing Data Processor under Article 4(8) UK GDPR, operating under a Zero Standing Privileges architecture. The standing Data Processing Agreement is operative continuously from signature; audit rights, breach notification, Article 32 technical and organisational measures, confidentiality, and liability apply at all times. In the Default Operating State, Private AI Ltd holds no active connection, no credentials, and no means of accessing Client systems — but its processor status is continuous. Active processing is limited to authorised Just-In-Time Support Sessions and Professional Services Engagements conducted under a signed Statement of Work, governed by the tiered instruction model in Schedule 2 of the standing DPA.

After handover, Private AI provides ongoing managed support without requiring access to your live environment. This includes continuous CVE monitoring against your Attested Software Bill of Materials (with 4-hour notification for Critical/Zero-Day vulnerabilities), monthly update scripts covering security patches and model weight updates (all SHA-256 verified), and emergency patch delivery within 2 working days. Technical support is available via two models: JIT (Just-In-Time) where our engineer connects via a time-limited SSH session with your explicit permission, or OTS (Over-the-Shoulder) where we guide your team verbally via video call. All JIT terminal sessions are fully logged and retained for 12 months.