Is it safe? AI, client data and a simple plan for your firm.
Ensuring AI safety in accounting firms hinges on a well-structured plan, proper settings, and thorough training. By understanding key terms and establishing clear policies, firms can protect client data while leveraging AI effectively.
Part of our AI in accounting coverage. See the full AI for accountants guide →
AI safety depends on the plan your team uses, how the settings are configured and how well people are trained. A meeting, a training session and a simple policy will get most firms there.
I speak about AI at events all over. There’s rarely a stage I step off without someone asking me whether AI is safe.
My honest answer is that the question is too broad. Here’s what I ask firms to look at instead.
Safety depends on the plan, the settings and the person
Think about a power saw. It comes with a guard for a reason, and it’s safe in trained hands. AI tools work the same way. The tool’s security matters, and so do the plan you’re on, the settings and how well the person using it understands what they’re doing.
Most worries about AI fall into four groups. Confidentiality is about where client data goes and who can see it. Accuracy is about AI being confidently wrong, which is why checking sources matters. Compliance covers the Privacy Act, your professional body’s rules and your insurer, who I’d call to make sure your policy reflects your AI use. Trust is about how clients feel, and that is changing as clients use AI in their own lives.
The plan your team logs into matters most
Whatever tool you use, know whether each person is on a consumer, business or enterprise plan. The brand might be the same, though the data terms are different.
On my reading of current terms, personal versions of some major AI tools train on your chats by default, while business and enterprise plans switch training off. Training means the provider can use your conversations to improve future models. Terms change quickly, so always check your own plan.
Ask your team which plan they’re logged into, then move all work onto a business or enterprise plan with work logins. That gives the firm admin controls, including removing access when someone leaves.
On a business plan, check the privacy, memory and connector permission settings. Connected tools can be set so each action is allowed automatically, needs your approval or is blocked. Rating a response with a thumbs up or down can send that chat to the provider as feedback, so I give feedback inside the chat instead.
Five terms to ask every vendor about
Accountants train in accounting, and vendor security jargon can be hard to follow. These five terms are worth understanding.
SOC 2 Type 2
SOC 2 Type 2 means an independent auditor has watched how a company handles data over six to 12 months. Ask what period it covers and whether you can see the report. A claim of compliance with no report to show is a red flag.
Sub-processor
A sub-processor is a third party your vendor passes data to. Ask who it is and whether that’s published.
Data residency
Data residency is the country where your data is stored and processed. Under the Australian Privacy Principles, sending personal information overseas means taking reasonable steps to make sure it’s protected.
Data processing agreement (DPA)
A data processing agreement, or DPA, is the contract that says the vendor only uses your data as you instruct, tells you about any breach and deletes data on request. I keep the DPA for every tool I use in one file.
Prompt injection
Prompt injection is when hidden instructions in an email, document or website try to get an AI tool to do something malicious. Ask vendors how their tool handles it.
When a client connects Xero to AI
More clients are connecting Xero to AI tools and then emailing their accountant about it.
When that email lands, ask which AI tool they’re using and whether it’s a business plan. Check whether they’re using Xero’s own connector or a third-party app, and whether it can change data. Find out who else has access to their AI account, and whether they expect you to rely on its output.
On your side, agree who checks AI output before anyone relies on it, and add a line about client-connected AI to your engagement letter. Checking AI-generated reports takes time, and that time should be billable. Keep client files out of your own AI tools unless the firm has approved it.
Shadow AI is your team asking for help
A KPMG study found more than half of respondents had used AI in ways that went against company policy. That’s shadow AI. I see it as a sign the team needs support with the tools they use.
The easiest way to find out what’s in use is an amnesty where nobody gets in trouble. You can’t control something you don’t know about. Admin centre usage reports, a list of AI domains from your IT provider and expense claims can fill the gaps.
Keep the basics strong too. Simple scams still work, like a caller posing as IT and asking for a password. For voice notes or video calls asking to change bank details, agree a safe phrase with clients and verify through a second channel.
A simple three-step plan for your firm
Step 1: Hold a meeting
Start with the amnesty, agree the approved tools and be clear about what data can go into them. Uncertainty about what’s allowed is one of the biggest barriers to adoption I see.
Step 2: Run a training session
A lunch and learn works well. Name an AI lead the team can go to, and agree a breach plan for the first 72 hours if something goes wrong.
Step 3: Hand out a guide and policy
Cover approved tools, data rules, plans and logins, vendor checks and where your DPAs are kept. Record where human review happens, and include AI access in your offboarding process.
Human review matters most. The TPB’s guidance is clear that AI doesn’t reduce or move your responsibilities. It expects evidence that someone has read, understood and checked what AI produced.
AI won’t go to jail for you. A clear plan means you’ll never need it to.
Is it safe? AI, client data and a simple plan for your firm
Billie Mcloughlin, Founder of AIccountancy, sets out what to check across AI plans, settings and vendors, with a three-step plan for safe adoption and a free shadow AI amnesty survey for your team.
Part of The AI in Practice Summit, 11 and 12 November 2026. All session recordings now available.
Get the editable AI client data rule sheet Download the AI client data rule sheet
The Word document (.docx), ready to fill in. Your profile unlocks every template on The Firm.
Or build it with AI
Open Claude or ChatGPT with the prompt filled in. It writes the finished document for your firm straight away, then asks a few questions to tailor it. No client data needed.
Works without client data. Don't paste client names, tax file numbers or ledgers into an AI tool your firm hasn't approved for them. AI data safety playbook.