Sovereign AI for Accountants: Confidentiality by Design

Sovereign AI lets an accountancy or audit firm use capable AI without sending a single client record to an outside cloud model. Mickai is a SIOS that runs on the firm's own hardware, owned, offline and provable, so tax files, ledgers and working papers never leave the practice. Every action is written to an Open Audit Record, giving partners tamper-evident proof of what the system read and produced. Confidentiality is preserved by design, not by a supplier's promise.

• Client data stays on the firm's own hardware, owned and offline, so nothing is exposed to a third-party cloud model.
• The professional duty of client confidentiality under codes such as the ICAEW and ACCA ethics codes is honoured by architecture, not by policy alone.
• Every prompt, retrieval and output is recorded to the Open Audit Record, supporting audit independence and record-keeping.
• Sensitive financial, tax and market-moving information never crosses an external network boundary.
• The Mickai closed beta is open to selected design-partner firms at mickai.co.uk/beta.

Why can't accountants simply use a public cloud AI?

Public cloud AI services process prompts on infrastructure the firm does not own or control. For an accountancy or audit practice, that means client ledgers, draft accounts, tax computations and market-sensitive deal information could be transmitted to, and processed by, a third party. Even where a provider promises not to train on the data, the firm has handed over custody and can no longer prove where the information travelled or who could reach it.

The professional duty of confidentiality does not soften because a tool is convenient. Regulators and clients expect a firm to keep control of the information entrusted to it. Sending that information to an external model, however capable, is a custody decision that many partners are simply not willing to make on behalf of their clients.

What does sovereign AI actually mean for a firm?

Sovereign AI means the firm owns both the intelligence and the data it runs on. Mickai is a SIOS, a Sovereign Intelligence Operating System that installs on the firm's own machines and runs entirely offline. There is no external inference endpoint, no shared tenancy and no telemetry leaving the network. The model runs where the data already lives.

Because the system is owned rather than rented, the firm decides what it processes, when and for whom. That ownership is what turns confidentiality from a contractual hope into a structural fact. The approach is backed by 104 filed UK patent applications, approximately 2,340 claims, owned by Mickai LTD.

How does running on your own hardware protect client data?

When AI runs on hardware the firm controls, client information is never placed in the hands of an outside processor. Working papers, payroll, VAT records and forensic material stay inside the same trust boundary the firm already maintains for its files. There is no network path for that data to reach a public model, because no such path is ever opened.

This also simplifies the firm's own obligations. Data that never leaves the building is far easier to govern, retain and, when required, delete. Rather than tracing a chain of sub-processors and hoping each one holds, the firm can simply demonstrate custody of the information from start to finish.

What is the Open Audit Record and why does it matter for independence?

The Open Audit Record is a tamper-evident log of everything the system does: the prompts it received, the documents it read and the outputs it produced. For an audit practice, that record supports independence and quality review, because the work an assistant contributed can be inspected directly rather than assumed after the fact.

Auditors are held to high standards of documentation and professional scepticism. A provable record of every AI-assisted step lets a firm show exactly how a conclusion was reached and keep the human judgement of the engagement partner clearly on top. It is evidence for the file, not a marketing claim.

How can a firm start with Mickai?

Mickai runs the Mickai SIOS on the firm's own hardware, owned, offline and provable, and we write every action to the Open Audit Record. The closed beta is open now and it is selective: we work with a small number of design-partner firms rather than accepting everyone, so that each deployment is shaped around the way real practices actually work.

Firms interested in confidentiality by design can apply at mickai.co.uk/beta. Mickai was founded by Micky Irons to give regulated professions the full benefit of AI without the loss of custody, and the beta is where we prove that approach alongside working accountants and auditors.

Frequently asked questions

Will my client data ever be used to train an external model?

No. Mickai runs on the firm's own hardware and offline, so client data is never transmitted to an external model and cannot be used for third-party training. The information stays inside the firm's own trust boundary at all times.

Does this help me meet my confidentiality obligations under the ethics codes?

It supports them directly. Codes such as the ICAEW and ACCA ethics codes require a firm to protect client information, and keeping that information on owned, offline hardware honours the duty by design. This is not legal, audit or tax advice, and each firm should confirm its own compliance position.

Can I prove to a reviewer what the AI did on an engagement?

Yes. Every action is written to the Open Audit Record, a tamper-evident log that a quality reviewer or regulator can inspect. That record supports audit independence, documentation and sound record-keeping.

Is Mickai just another cloud application?

No. Mickai is a SIOS that the firm owns and runs on its own machines, not a hosted application operated by someone else. There is no external inference endpoint, so nothing about an engagement depends on an outside service staying online.

How do I apply for the beta?

Firms can apply at mickai.co.uk/beta. The beta is selective and works with design-partner firms, so an application is the start of a conversation about fit rather than a guaranteed place.