When AI tools go wrong: Who wears the risk?

Artificial intelligence (AI) is rapidly finding its way into accounting practices. From drafting emails to automating repetitive tasks, many firms are experimenting with AI tools to improve efficiency. A newer development is the rise of “vibe coding” where users build small applications or agents using AI prompts rather than traditional coding. 
​
While the technology is impressive, it raises an important question for accountants. If an AI-built application mishandles client data or causes a loss, how does that interact with professional indemnity (PI) insurance? 

Where responsibility is likely to sit
From a professional indemnity perspective, the core principle remains unchanged. A PI claim responds where an accountant has breached their professional duty and a third party has suffered a financial loss as a result.

If an accountant chooses to use AI to build or deploy an application and that decision leads to client data being exposed, misused or relied on incorrectly, the liability is unlikely to rest with the AI provider alone. In many cases,
responsibility would sit with the accountant who decided how and where the tool was used, what data it could access, and how it was rolled out.

This is particularly relevant where client information is uploaded, connected, or made accessible to systems that were not designed or approved for that purpose.

AI mistakes are not a surprise 
There is an important distinction emerging in this space. 
Using an AI tool is one thing. Publishing or deploying an AI agent that interacts with systems, data or workflows across a firm is another. 

Simple tools used by individuals for drafting or idea generation generally present a different risk profile to agents that can read files, process data, trigger actions or integrate with other platforms. As agents become more capable, the consequences of design errors or poor controls become more serious. 

PI and technology 
PI insurance does not insure technology itself. It responds to alleged failures in professional services. 

If an AI-related issue results in an allegation that an accountant failed to protect client data, failed to exercise due care, or relied on outputs that should not reasonably have been relied upon, that is where PI exposure may arise. 

This makes risk management around AI use critical. 

Practical risk management steps 
Accounting firms do not need to avoid AI completely. However, sensible controls matter. 

Limit publishing, not experimentation 
Allow individuals to build or test simple agents that make repetitive work easier, but avoid firm-wide deployment without oversight. When an agent is published across a practice, it should be reviewed in the same way as any other system or process change. 

Be clear on access and permissions 
Before rolling out any AI agent, understand exactly what it can access and what it can do. Key questions include what data it can read, what actions it can take, and what external content it connects to or collects. 

Update your AI policy 
Many existing AI policies focus on tools rather than agents. Agents raise different questions, including who can create them, who can approve publication, how they are tested, and what happens if something goes wrong. Policies should evolve as the technology evolves. 

Avoid client data where possible 
Where practical, keep client data out of experimental or untested AI systems. The fewer touch-points between AI and sensitive data, the lower the professional and privacy risk. 

An evolving conversation 
AI in accounting is still developing. The technology is moving quickly and professional standards, client expectations and regulatory guidance are still catching up. What is acceptable today may look very different in a year. 

For most practitioners, the key takeaway is not that AI should never be used. It is that using AI carries the same professional responsibilities as any other tool or system. Thoughtful controls and clear governance help reduce the risk that an efficiency experiment turns into a professional indemnity issue. 

AI does not change the fundamentals of professional responsibility, but it does introduce new ways that things can go wrong. Reviewing your professional indemnity cover alongside your firm’s approach to technology, data and governance can help ensure you are protected if a claim arises. 

Want an obligation-free quote? 
We specialise in providing professional indemnity insurance to Australian accounting professionals. Depending on your circumstances, you could benefit from Accountancy Insurance’s PI Shield scheme. Otherwise, a tailored professional indemnity insurance policy can be created specifically for your needs. 

• Accountants, click here. 
• Bookkeepers/BAS Agents, click here. 
• Tax Agents, click here. 

A huge thanks to Karen McDonald from accountancy insurance for providing us with this tailored article on AI in accounting. Want to learn more? Don’t hesitate to reach out to the team at accountancy insurance.