Even the most judicious users of AI in advice could find their standards slipping without the right policies in place, according to Marshan Consulting director Ben Marshan.
Referencing research from the Financial Planning Standards Board from April 2025, Marshan noted that around two-thirds (64 per cent) of advice practices were using or planning to use AI – 38 per cent of which were currently using it, 16 per cent were piloting it and 10 per cent aimed to integrate it over the next 12 months.
However, only 45 per cent had an AI policy in place, and just 17 per cent had what the research described as a comprehensive policy.
“So, roughly 50 per cent of the planners who responded to this survey were using AI without a comprehensive plan, without a strategy for how you use it without rules and governance in place around what they were putting in,” Marshan said.
To illustrate the dangers of this approach, Marshan used the case study of a sole practitioner working in a “rural-ish” area.
“The licensee has no AI policy, but she decides she’s going to use AI and see how it could help her practice. She has an appointment with two new clients, Jake and Sam, a couple in their early 40s who’ve just sold their café,” Marshan said.
Marshan explained that the couple came in with a fairly routine question about having enough for retirement. The adviser discloses she’s going to use AI upfront and that she’s going to protect her clients’ data.
“She’s taken her clients’ names out of any information that she’s put into them. She’s got paid accounts so that the information that she puts in into the projects that she’s using and not used for training purposes,” Marshan continued.
“She pressure-tests the AI models against her own professional view, so she’s tried a couple of different ones, figured out which one gets closest to the answer, and puts them through their paces to make sure that she can rely on the outputs that come out. She’s used AI as an assistant, but she has advised the client.
“So, the competence, the output, the advice she provides is hers.”
Over time, Marshan said, “word got out” about how AI helped this adviser with her clients. And because of this, she managed to onboard 15 new clients.
“The licensee still hasn’t released a policy at this point,” Marshan said.
“Her policy is still in draft; it’s not updated. It was last open six months ago, when she first drafted it. She hasn’t done anything wrong since then, but there has been an erosion of how she has used AI in her practice.”
Marshan said this “erosion” manifests in small ways: the adviser started copying client information directly into the AI tools without anonymising it first, for example. She also started accepting the outputs of her modelling tools without verifying them.
“She’s just waving them through because she has gotten so busy with the work that she’s doing. She hasn’t stayed competent. She’s less learning on her own and more relying on the tool to help her do the work,” Marshan said.
In fact, Marshan said, the adviser had become so confident in her use of AI that she used those first two clients – Jake and Sam – as a case study for social media. She used AI to generate the posts.
“Given that she lives in a small community, in a small regional area, with a population that knows her and probably knows other people that live in that community, there’s enough information there within her community that they can start to identify Sam and Jake,” Marshan said.
“It’s been a good outcome, but the information that’s in there is enough to start to identify them. So, the confidence that she’s got is starting to outrun that care for the client that she should have as a professional.”
Marshan said these kinds of scenarios aren’t uncommon, and that there’s a “lot of risk sitting there” if advisers don’t fully understand the systems they’re using.
Plus, he said, ASIC has taken a keen interest in this area. Last month, the regulator issued an open letter to AFSLs with a clear message: “Do not wait for perfect clarity to address the threat posed by new AI models. Instead, act now, and act with discipline.”
“They’re deadly serious about the concern they have over some of these new frontier models that are about to hit and the cybersecurity risk that will come,” Marshan said.
He added: “My point here is that they’re very clearly saying that the clock’s moved. It’s a minute to midnight. These tools are about to come out. You need to understand how secure your data is and how secure your systems are, because these AI tools are coming.”
