TPB’s draft AI guidance needs greater practical utility: CA ANZ

In its submissions, CA ANZ listed five recommendations to provide targeted improvements to the draft, ensuring clear, actionable, and relevant guidance for all practices, regardless of their size.

“.... a number of targeted additions would materially improve the guidance’s practical utility, particularly for small and micro practices that may not have dedicated technology, privacy or cyber security teams,” the body said.

CA ANZ’s recommendations stressed that the guidance must go into more detail regarding client confidentiality and privacy, managing the risks of using third-party AI tools, and providing clarity around privacy with the use of internal AI tools vs third-party tools, and including worked examples of AI-use cases for low, medium, and high risk applications of the tool.

It recommended that the TPB publish a checklist of privacy matters that should be disclosed to clients in its guidance, such as unauthorised disclosures, such as entering client TFNs into public LLMs. In addition, it recommended that examples be provided to help practitioners obtain informed permission from clients regarding information that may be disclosed to an AI provider and the use of AI features embedded in internal software.

Further, CA ANZ called for the draft to be amended to distinguish how it deals with external and internal AI tools, as “they offer vastly different levels of data privacy, accuracy, workflow integration, reliability, and accountability,” the submissions read.

In addition, the body recommended that examples referencing case law be included in the guidance to illustrate real-world applications and “provide more operational guidance on what reasonable controls look like in different practice contexts”. It said that this short set of worked examples will show how the code obligations apply to common scenarios for low-risk, medium-risk, and high-risk AI use-cases.

CA ANZ also recommended that staff be required to reference primary materials such as case law from reputable websites, and avoid relying on AI-generated citations unless independently confirmed.

“The TPB should ensure the final version clearly states that the guidance is principles-based and that practices should apply proportionate controls having regard to their size, services and client base,” CA ANZ said.

In its submissions to TPB’s draft guidance, CPA Australia also urged that the draft guidance advise tax practitioners to verify and review AI-generated content for accuracy, and establish processes to understand and contest AI outputs, emphasising that AI may not always be accurate or factually correct.

The consultation period for the draft guidance ended on 21 April, with the board now considering submissions.