The Cybersecurity Case for On-Premise AI, Not Just the Compliance One
Published 28 August 2026
Short answer: Every cloud AI tool your firm uses is a new external system holding a copy of your client data: a new vendor with its own staff, its own breach history, and its own attack surface, outside your control. The compliance guidance (TPB, QLS, AHPRA) explains why regulators care. The plainer reason to care yourself is that a vendor holding your clients' data can be breached whether or not you did anything wrong.
Most of the conversation around AI and client confidentiality gets framed as a compliance question: what does the TPB require, what does QLS expect. That framing is accurate but incomplete. Underneath it is a much older, simpler question: every additional place your data sits is a place it can be stolen from.
The uncomfortable math of third-party data
When a firm sends client data to a cloud AI tool, it isn't just trusting the AI model. It's trusting that vendor's entire security posture: how they store data, who on their staff can access it, how they patch vulnerabilities, how they respond to a breach, and how many other companies' data sits on the same infrastructure as a shared point of failure.
None of that is visible from the outside. A firm can read a vendor's privacy policy and security page, but it can't audit their internal access controls, and it has no control over the vendor's own mistakes. AI vendors, like any software company, have had security incidents, and the more AI tools a firm plugs into its workflow, the more of these external dependencies exist.
Why professional services firms are a specific target
Accounting practices and law firms sit on exactly the data attackers value: identity information, financial account details, trust structures, litigation strategy, unreleased financial results. A breach at a single AI vendor doesn't just expose one firm's data. It can expose every client of every firm using that vendor at once, which is precisely what makes a vendor worth attacking.
That's a different risk profile to a firm's own systems. A firm controls its own staff training, access logs, and incident response. Once client data goes to a third-party AI tool, that control passes to a company the firm has never audited and generally can't.
Where this meets the regulatory guidance
TPB(GS) 55/2026, QLS Guidance Statement No. 37, and AHPRA's AI guidance are written for different sectors, but they converge on the same instinct: once client data leaves your systems, you've lost the ability to fully control what happens to it. The regulators are naming a security reality, not inventing a bureaucratic hurdle. "Where does the data go" is the compliance question and the cybersecurity question at the same time.
That also means fixing one tends to fix the other. A firm that can honestly say client data never leaves its own network has answered the disclosure question the TPB and QLS are asking, and removed an entire category of breach risk that has nothing to do with either regulator.
What this looks like in practice
- Fewer external dependencies: every cloud AI vendor removed from the workflow is one less company whose security your clients' data depends on.
- A smaller, known attack surface: client data stays on hardware your firm owns, inside a network your firm (or your IT provider) already secures.
- No sub-processor chain: cloud AI products often rely on other companies behind the scenes for hosting, analytics, or model infrastructure. Each one is another party with some form of access. On-premise AI has none, because there's no third party in the first place.
The question worth asking before the compliance question
Before asking whether a specific AI tool satisfies Code Item 6 or ASCR confidentiality obligations, ask the simpler version: if this vendor were breached tomorrow, whose data would be in the headline, and would your clients have to find out it was theirs? For firms whose entire value rests on discretion, that's not a rhetorical question.
See how Althorn's on-premise AI works
Book a 20-minute confidentiality audit, no obligation.
Book a 20-Minute AuditCommon questions
Is cloud AI safe for confidential client data?
It depends on the vendor and the sensitivity of the data, but every cloud AI tool adds an external system holding a copy of client data, with its own staff, security posture, and breach risk outside your control.
Why are accounting and law firms a particular target?
They hold identity documents, financial details, trust structures, and litigation strategy. A breach at one AI vendor can expose every client of every firm using it.
How does on-premise AI reduce breach risk?
Client data stays on hardware your firm owns, inside your own network, with no external vendor or sub-processor chain holding a copy.