AI In Accounting: What Are We Actually Exposing?
The Question Every Principal Should Ask
Before we talk platforms, regulation or future compliance, there’s a simple test every principal should apply:
“Would my client be comfortable knowing this information was entered into an AI system?”
If the answer is no — or even maybe — then you have a risk exposure. Not a theoretical one. A practical, reputational, and potentially legal one.
AI doesn’t need to “leak” data for it to be a problem. It only needs to retain, learn from, or be accessed in ways you didn’t fully understand when you uploaded it.
Not All AI Tools Are The Same — And That’s Critical
One of the biggest misconceptions I see is the belief that all AI tools treat data the same way.
They don’t.
There is a meaningful difference between enterprise-grade tools with contractual data protections, AI embedded inside regulated platforms, consumer-facing AI tools, and the small “helpers” bolted onto existing software.
Some are ring-fenced at an organisational level, with explicit commitments not to train on customer data, administrative controls in place, and the ability to monitor usage. Others may retain prompts, use inputs to improve their models, store information offshore, or operate under terms that shift over time.
For example, enterprise versions of popular platforms — such as Microsoft Copilot for Microsoft 365 or Google Workspace tools — typically include contractual commitments not to train on customer data and provide administrative controls. Standard consumer versions of the same underlying models may not.
If you don’t know which category your tools fall into, that’s the gap.
“Would my client be comfortable knowing this information was entered into an AI system?”
The Real Risk: Unstructured Use Inside Practices
In most firms, the biggest exposure isn’t AI policy — it’s behaviour.
It shows up quietly. A client email pasted in “just to tidy it up.” Financials uploaded for explanation. Draft advice refined outside core systems. Notes from sensitive meetings summarised to save time.
None of this is malicious.
All of it is understandable.
But without clear boundaries, the practice — not the individual — carries the risk.
What Sensible Practices Are Doing Right Now
The more thoughtful firms aren’t banning AI. They’re governing it.
They are clear about boundaries — what can be entered, what must never be entered, and which tools are approved. Not in a 30-page legal document no one reads, but in practical guidance that people actually follow.
They treat platforms differently, rather than assuming “AI” is one uniform category. They define which tools are approved, what level of data can be used in each, and who has authority to introduce new systems.
Where possible, they use enterprise versions, disable training on proprietary data, control access, and ensure activity is auditable.
And importantly, they educate their teams. Not with fear-based messaging, but with awareness. What AI does well. Where it introduces risk. When to pause before pasting something in.
You don’t need everyone to be a technologist.
You need them to be conscious.
AI Use & Data Security
Internal Standard Operating Procedure
Purpose
This SOP sets boundaries for the safe and appropriate use of AI tools within the practice to protect client confidentiality, proprietary information, and professional obligations.
Approved Use
- 1 Drafting generic emails or documents
- 2 Summarising de-identified information
- 3 Improving grammar, structure, or tone
- 4 Brainstorming using non-confidential inputs
- 5 Working with public or approved templates
Prohibited Use
- 1 Client identifiers or personal information
- 2 Financial statements or tax data
- 3 Deal structures, valuations, or pricing logic
- 4 Proprietary methodologies or advisory IP
What Sensible Practices Are Doing Right Now
Client data is only part of the picture.
AI also touches proprietary methodologies, pricing logic, internal models, strategic IP, deal structures, succession frameworks, and valuation approaches.
Even anonymised or aggregated information can contribute to model improvement in ways that gradually erode a firm’s competitive edge.
For advisory-led firms, that is not a small issue.
It goes directly to enterprise value.
The Regulatory Direction Is Clear — Even If The Rules Aren’t (Yet)
In the health sector, there are already professional and privacy guidance expectations around transparency and consent when AI influences patient-facing decisions. Bodies such as the Australian Health Practitioner Regulation Agency have been moving steadily toward clearer expectations around digital tools and data handling.
Accounting is not there yet.
But it would be naïve to assume it won’t follow.
The Office of the Australian Information Commissioner has already made it clear that organisations remain responsible for personal information even when third-party tools are involved. That principle does not disappear simply because AI is involved.
It is entirely plausible that in future we see expectations around disclosure of AI use, client consent, professional standards referencing AI governance, and insurers asking sharper questions.
The firms that get ahead of this will not be scrambling later.
The Bigger Strategic Question
AI is not going away, and realistically, nor should it.
Used properly, it improves efficiency. It sharpens thinking. It reduces repetition and manual effort. Most firms are already benefiting from that.
But benefit does not remove responsibility.
The real question for principals is not whether AI should be used. It is whether the practice understands what information is being shared, where it is going, and under what controls.
If that answer is unclear, the issue isn’t technology.
It’s governance.
Accounting is not there yet.
But it would be naïve to assume it won’t follow.
The Office of the Australian Information Commissioner has already made it clear that organisations remain responsible for personal information even when third-party tools are involved. That principle does not disappear simply because AI is involved.
It is entirely plausible that in future we see expectations around disclosure of AI use, client consent, professional standards referencing AI governance, and insurers asking sharper questions.
The firms that get ahead of this will not be scrambling later.
-
Typical Fees for Common Services Report
Price range: $198 through $396 incl. GST This product has multiple variants. The options may be chosen on the product page
Preparation Beats Panic — Again
We’ve been here before. Privacy reforms tightened. AML/CTF obligations expanded. Cyber security expectations increased. Professional indemnity scrutiny followed.
Each time, the firms that acted early had far less disruption than those that waited.
AI will follow the same path.
The practices that set boundaries now, document their decisions and educate their teams will be far more comfortable in a few years’ time than those who postpone it.
Because eventually, these questions won’t feel abstract. They’ll feel immediate.
And by then, hindsight won’t be helpful.
Written by Catherine York
Catherine York works alongside Rob Knights & Co on governance, systems and digital strategy. A former ASIC investigator with over a decade of experience in regulatory environments, she specialises in operational risk, compliance frameworks and technology implementation for professional service firms.
What practices are selling for, and what is moving the numbers
Michelle’s monthly note on practice valuations, buyer activity and the conditions we are seeing across accounting, financial planning and legal firms.
