Section 10.37 turns an AI tool's untraceable logic into a cost tax firms pay twice

The cost of an untraceable tool shows up twice in the same engagement. Kevin Boeckholt, a CPA at Accordance, wrote in CPA Practice Advisor that when a tool cannot produce a source trail, the firm builds one by hand ^[10][12]. That means re-deriving citations, re-checking calculations, and cross-referencing regulations the AI already claimed to have relied on ^[10]. He described a preparer using an opaque tool as "paying twice for the same research: once to generate it, and once to confirm it actually happened the way the output implies" ^[16]. Those verification hours are real hours, so a firm can bill them, while the alert reads 10.27(a) to mean it cannot bill the manual research time the tool replaced ^[2]. So either the client's bill does not fall, or the licence comes out of the firm's margin ^[2][10].

Sort the alert by provision and the weight sits on verification. Of the four sections the column works through, three go to what the practitioner can independently confirm: 10.37 on written advice, 10.22 on due diligence, 10.35 on competence. The fourth, 10.27(a), is about the invoice ^[14]. On competence, Boeckholt wrote that understanding how a system generates content "cannot mean reading a vendor's marketing page or trusting a benchmark score" ^[13].

Deloitte Australia's 237-page report for the Australian government, published in July 2025, contained invented quotes attributed to a judge and citations to academic works that did not exist, apparently generated by AI and unreviewed before delivery ^[5]. An outside academic checked the citations against the sources and found that some of them did not exist at all ^[7]. Deloitte Australia refunded part of its fee, and the column does not state the amount ^[6]. OPR's alert came about eleven months after the report ^[15].

The case for the fee provision being the live risk is about what an examiner can see. Hours not incurred sit in a firm's own time records, whereas unreasonable reliance on written advice under 10.37 needs a wrong answer plus someone motivated to trace it back to a source ^[3][7]. That asymmetry is why I would expect the first enforcement noise to be about invoices even if the durable exposure is the citation trail.

In my view the fee debate is downstream. Boeckholt put the standard this way: "Opacity, not cost and not billing model, is the standard tax practitioners now need to apply when they evaluate an AI tool" ^[11]. He argues the billing question cannot actually be answered until the traceability question is resolved ^[17]. Evidence against that reading would be OPR's first published AI matters arriving as fee disputes under 10.27(a), with the traceability language left as guidance nobody has to price.