This article is the first in our special report on digital risk in family offices.
By now we’ve all heard of lawyers who lost their jobs after citing nonexistent case law, and that reviewer who recommended imaginary books after using generative AI tools that “hallucinated” supposed facts. But we’re still in the early days of understanding the risks associated with employing AI for professional purposes—especially those that might accrue from the actions of an associate or service provider.
Of the many potential pitfalls for family-office advisors, “privacy is probably number one,” says Neil Cockburn, director of Family Enterprise Advisory Services with KRP (Kingston Ross Pasnak LLP), based in Edmonton.
Cockburn is active among his regional FEA (Family Enterprise Advisor) peer group, where this topic “is very top-of-mind,” he says.
A particular concern is how to protect client data handled by outside firms that use AI tools. “In our last session, the topic of contracting came up,” says Cockburn.
Within KRP’s Family Enterprise area, “we’re reviewing our engagement letters across the board, and within our contracts we’re adding clauses with respect to use of AI, in privacy and other issues.”
A lot of systems are ‘powered by AI,’ without a lot of explanation to the user about where the data is touching AI.
Ryan Clarke, Edmonton-based tax advisor
AI clauses in contracts are becoming more common, says Vancouver lawyer Jillian Frank, who is the National Lead, Legal Transformation, Technology and Managed Services for KPMG Law Canada. This is partly because contracting firms have been reviewing their potential risks in engaging providers, and partly because providers are protecting themselves from liability.
“Some of the courts are requiring disclosure of AI in legal submissions, and the Office of the Superintendent of Financial Institutions has been asking for disclosure as well,” says Frank.
She identifies four main areas of AI risk:
- Bias and inaccuracy of AI-created content.
- Exposure of private data stored in AI-based tools.
- The related issue of leaking confidential corporate information or intellectual property.
- The potential for AI-generated outputs to violate laws.
“Summarizing, brainstorming, but not relying on the output as a source of truth: those are green-light, low-risk activities,” says Frank. A higher level of risk is associated with such tasks as issue-spotting or comparing contracts, where “you need to have enough professional knowledge to know ‘oh, that’s wrong!’”
The next stage is “when you’re quoting exact legislation or coming to a final legal decision,” she says. “You need someone looking through that who has not only the knowledge but also the experience on how that has been used over past decades; that’s where you have to be really cautious when bringing in that outside person.”
Knowing whether AI was used doesn’t tell you whether it was used well.
Eli Bernholtz, founder of Virtual Family Offices
Among other issues, firms that fail to examine how personal or corporate information is being handled, stored and secured may be in contravention of privacy legislation. “If it’s something that you wouldn’t want to expose to a competitor or your next-door neighbour, you certainly shouldn’t be putting it into an AI tool,” Frank says.
Family offices are increasingly requiring disclosure, with larger, institutional providers more likely to proactively disclose, while others respond when asked, says Jaz Gill, the Burlington, Ont.-based CCO and SVP, Technology and Operations, with Northwood Family Office in Toronto.
“In practice, disclosure is expected where AI is directly involved in client interactions or outputs,” he says. “Sophisticated clients are beginning to raise this explicitly as part of due diligence.”
Contracts are beginning to include AI conditions such as “restrictions on uploading confidential data into third-party AI tools and clarifying how data can be used or stored,” says Gill. However, “the trend is moving in a direction of clearer, more specific language rather than outright prohibitions.”
We place a lot of trust in our service providers to ensure that they’re using our data correctly.
Ryan Clarke, Edmonton-based tax advisor
Eli Bernholtz, based in Montreal, is the founder of Virtual Family Offices (VFO). “My business uses an outsource model, so we don’t have the in-house lawyers, investment advisors and accountants on payroll. As a result, I meet with all these experts from other firms who are using these tools for their data operations,” he says.
“What is important, for us, is for the privacy and confidentiality of the families, that we actually understand how these professionals are using this data to train AI,” Bernholtz says.
“When you’re using an AI model, you’re using an outside platform that is beyond your control. It will give you analyses and really good information on what you can do with it, but you’ve given it access to highly confidential data. Disclosure is an important starting point, but knowing whether AI was used doesn’t tell you whether it was used well.”
Bernholtz emphasizes that clients must be involved in making decisions about AI use, and he cautions that the generations approach it differently: “It can’t be the older generation, top-down, telling people we can’t use AI. These conversations have to be had really early on.”
Ryan Clarke, who is based in Edmonton, has his own practice as a Canadian tax advisor and also serves as an advisor within Winnipeg-based Blackwood Family Enterprise Services. He says there’s no point in asking whether your associates are using AI—everyone is. “The better question is: Where is AI touching confidential family information?”
He points out that, apart from popular tools such as ChatGPT, Claude and Copilot, “a lot of systems are ‘powered by AI,’ without a lot of explanation to the user about where the data is touching AI. We place a lot of trust in our service providers to ensure that they’re using our data correctly.”
The onus is on advisors to think about accountability.
In both of Clarke’s practices, “what I am seeing now is asking for targeted disclosure. I have a privacy policy on the website for my tax-advisor practice stating that I may use AI for research and that all care is taken to avoid submitting personal information into AI tools. It shows our clients that we have put some thought into it,” he says.
“We’re at a unique time right now with information systems; it would be a mistake for family offices or their advisors to ignore it,” says Clarke. “The best practice right now is to treat AI like any other outsourced service that can affect confidential family information.”
Jillian Frank points out that while there’s a lot of discussion about safety and what professionals are going to do in the future, there’s also a lot of human creativity happening.
“I think it’s actually really exciting to work on these issues.”
Sarah B. Hood is a Toronto-based writer and book author. She has served as editor of three national magazines and written weekly columns for the National Post. She also serves on the editorial board of Spacing magazine. She writes frequently on business, urban affairs and culture. As a food writer, her work has been translated into Japanese and Arabic. She has taught writing at George Brown College for more than 20 years.
The Canadian Family Offices newsletter comes out on Sundays and Wednesdays. If you are interested in stories about Canadian enterprising families, family offices and the professionals who work with them, you can sign up for our free newsletter here.
Please visit here to see information about our standards of journalistic excellence.