IntelekIQ
Secure Law · Secure Accounting
Scenario paper  |  Enterprise cloud AI vs firm-boundary inference

Sealed Desktop, or Sealed Inference?

Citrix can control the desktop. Microsoft 365 Copilot can provide governed AI within Microsoft's service boundary. Secure Law and Secure Accounting address a separate question: how to process confidential client files on premises, under the firm's control.

For: managing partners, principals, DPOs, and IT · UK, Malaysia & Singapore and similar markets  ·  Reading time ~11 min  ·  SC1 · August 2026

Consider a familiar setup. Lawyers and staff work through Citrix. Microsoft 365 holds mail and files. The firm has approved Microsoft 365 Copilot under enterprise terms. Each part can have strong controls. The remaining question is where AI processes confidential client files.

Citrix and Microsoft 365 Copilot can each be a sound investment. Neither moves AI processing onto the firm's premises. The term “sealed” can obscure that distinction.

1Vendor assurance does not define the processing boundary

Large platforms are familiar to boards, auditors and IT teams. They offer mature security controls and substantial operational capability. Those are valid reasons to use them.

The vendor name does not answer where a model processes a prompt. The firm must still identify the service operator, processing location, contractual controls and applicable jurisdictions. This is an architecture and governance question, not a judgement about whether a vendor is safe.

Vendor assurance and processing location are separate questions. Professional firms should assess both.

2Three things collapsed into one word

Partners and IT often treat three different controls as if they were the same thing. They are not.

What people say What it usually means What it does not decide
“Sealed desktop” Citrix or VDI session control: where the user’s desktop runs Where an AI model reads the prompt
“Enterprise Copilot” Contractual and tenant protections inside Microsoft’s cloud Whether inference is inside the firm’s own boundary
“Sealed inference” Firm-controlled processing for confidential, cited answers from firm files Whether mail and files must leave Microsoft 365

Where the prompt goes

Desktop control
Citrix / VDI

Locks how people reach the office. Does not move the AI model into the firm.

Enterprise cloud AI
Microsoft 365 Copilot

Processes prompts inside Microsoft’s service boundary under enterprise terms.

Firm-boundary inference
Secure Law / Secure Accounting

Designed so confidential, cited answers from firm files happen inside an architecture the firm controls.

Three controls. One word. Only the third answers custody at inference.

3Citrix controls the desktop, not the AI service

Citrix, used well, strengthens the endpoint story. Work can stay in a managed session rather than on an unmanaged laptop. Policy can be applied centrally. In some designs, local copying is harder. For many professional firms that is a material improvement in day-to-day control.

What Citrix does not do is relocate Microsoft 365 Copilot’s model into the firm’s building. Opening Copilot inside a Citrix window still sends the work into Microsoft’s service boundary. The desktop is managed. The inference layer is still Microsoft’s cloud AI, operating under the firm’s tenancy and Microsoft’s product terms.

Citrix may help control copying, local storage and access from unmanaged devices. It does not determine where an AI model processes client text.

4Microsoft 365 Copilot provides governed cloud AI

According to Microsoft’s public documentation for commercial Microsoft 365 Copilot, prompts and responses are treated as customer data under enterprise terms; they are not used to train foundation models; data is encrypted; tenants are isolated; Copilot respects the firm’s Microsoft 365 permissions; and Purview-style audit and retention tools can apply. Residency options exist for interaction content at rest. Those are real protections, and firms are right to take them seriously.

They are also a particular kind of protection: by contract, tenancy, and Microsoft’s controls. At the moment the model answers, confidential text is still processed in Microsoft’s cloud AI stack. Encryption at rest does not mean the model never sees readable text while generating an answer. Zero-training promises do not move the inference boundary into the firm.

This is the same distinction drawn in IntelekIQ’s cloud-versus-sealed papers: governed cloud AI and on-premises private AI use different processing boundaries. Both can be legitimate. They suit different work.

Enterprise Copilot can be the right tool for governed office AI. It is not, by that fact alone, firm-boundary custody for privileged matter files or regulated client numbers.

5One supplier can provide several different controls

The same vendor family may provide email, files, identity and AI. Security materials often discuss zero trust, encryption and compliance together. These controls matter, but they do not all operate at the same layer.

Precise language helps. When an environment is described as sealed, ask whether that refers to the desktop, the cloud tenancy or on-premises AI processing.

6The question that settles it

Use one publishable test:

At the moment the assistant reads our client’s tax file or matter memo, where is that data, and who operates the system that could be compelled to produce records of that processing?
  • Public chatbot: the provider’s cloud; the provider.
  • Microsoft 365 Copilot: Microsoft’s service boundary under the firm’s tenancy terms; Microsoft as operator of that layer.
  • Secure Law / Secure Accounting: on the firm's premises, under the firm’s control.

Microsoft provides substantial data protections. Operator jurisdiction and lawful access remain part of the architecture choice. Professional firms should consider them alongside security, contractual and operational controls.

7Coexistence, not replacement

Secure Law and Secure Accounting work alongside Citrix and Microsoft 365. The firm can keep its desktop controls, cloud storage and collaboration tools. It can use on-premises private AI when work involves confidential client files and requires answers from authorised firm material.

That is the same coexistence story as the rest of IntelekIQ’s positioning: practice systems and office platforms run the firm; private AI sits beside them for research, document review and drafting across authorised files. See also Control in the office.

8Who this paper is for

The primary readers are partners, principals, DPOs and IT leads in small and mid-size law and accounting practices. These firms need useful AI while meeting confidentiality and professional duties.

The same distinction can help larger organisations. Enterprise cloud AI may suit governed office tasks. On-premises private AI may suit confidential client work where the firm wants processing under its own control.

The distinction in one paragraph

Citrix can control the desktop. Microsoft can protect an enterprise Copilot tenancy. Neither, by itself, moves AI processing onto the firm's premises. Secure Law and Secure Accounting are designed for that separate role.

Further reading: Control in the office; Secure Cloud, or Sealed? (SLIA); The Assistant You Already Trust (SAIA). Product paths: Secure Law and Secure Accounting. Discovery: talk to us.

IntelekIQ · Scenario paper SC1 · August 2026 · Orientation for professional firms; not legal advice. Product names Secure Law (SLIA) and Secure Accounting (SAIA). No named clients. Enterprise Copilot descriptions follow Microsoft’s public commercial documentation as of the drafting date and should be re-checked against current product terms before reliance.