Cloud, On Premises or Private

1Three deployment models

AI processing can take place in a public or third-party cloud service, in a private service within the firm's cloud environment, or on premises on hardware controlled by the firm.

Each model can be appropriate for different work. The decision should reflect the sensitivity of the information, the provider's terms and controls, the firm's technical resources, and any client or regulatory requirements.

On-premises deployment is not automatically safer. A poorly maintained server exposed to the internet may present more risk than a well-managed cloud service. Hardware location does not replace patching, monitoring, access control or backup.

The useful question is what control the deployment gives the firm at the point of processing. Secure Accounting runs on premises so confidential client work can remain within the firm's controlled environment.

2How Secure Accounting separates the work

Secure Accounting separates private processing from internet-facing research. By default, research, document review and drafting involving confidential client work take place on premises.

External research can be carried out without including client identifiers. Verified, non-confidential material can then be added to the firm's approved knowledge library. KYC packs, facts relating to a suspicion and trust-account records remain outside the AI path.

When an external model is useful, optional Frontier Mode applies reversible pseudonymisation. Names and identifiers are replaced with placeholders, the accountant reviews what is sent, and original identifiers are restored within Secure Accounting. The accountant remains responsible for checking the result.

3The claim to make, and the one to avoid

The accurate product claim is specific: Secure Accounting runs on premises and keeps confidential client work within the firm's controlled environment by default.

This is not a claim that cloud services are insecure. Reputable cloud services may be well suited to many parts of the firm's technology estate. The reason to use private AI on premises is to give the firm direct control over processing for confidential client work.

If the firm uses Microsoft 365 or another cloud platform, its files may already be stored in that service. Secure Accounting does not make the firm air-gapped and should not be described that way. It changes where AI processing takes place for the work assigned to it.

4Choosing, in practice

The decision is rarely all cloud or all on premises.

General research, non-client drafting and learning may be suitable for an approved public or cloud AI service. Ledgers, tax files, audit evidence and other confidential client work need stronger controls. Secure Accounting provides an on-premises private AI route for that work. Excluded material, including KYC packs and facts relating to a suspicion, remains outside the AI path.

The classification scheme from the previous paper helps the firm decide which route applies. The firm can use different deployment models for different categories of work.

5The honest bounds

On-premises private AI still requires careful operation.

The firm needs suitable hardware, maintenance, access management and monitoring. Local models may also perform differently from leading external models on some tasks. Frontier Mode offers an optional route where an external model is needed, using reversible pseudonymisation and accountant approval before anything is sent.

Deployment on premises does not remove the firm's underlying duties. Confidentiality, anti-money-laundering obligations, data protection and breach notification continue to apply. Accountants remain accountable for the work and must review AI output.

Secure Accounting is a control within the firm's wider governance. It is not a substitute for that governance.

The next papers examine retrieval and client permissions (C5), the process for adding verified external research (C6), and AI over structured financial data (C7).