YOUR BUSINESS
What makes this more difficult is that errors in the higher risk category rarely show up immediately. They surface later— say when a client raises a concern or the firm reviews the record and finds that an omitted risk-tolerance note in a meeting summary led to a mistake that carried forward into a recommendation. The error might not arise until months later, when a client disputes a problem and no one else has caught it.
What Disciplined AI Use Actually Looks Like
You’ re not going to understand that gap— when a tool adds value and when it adds risk exposure— by looking at the tool itself. It’ s the way you approach it. It’ s what happens before the question gets asked.
AI doesn’ t create fiduciary responsibility, remove it or share it. It simply changes how decisions get made.
For instance, most people plug a question into a prompt and take the first response. A better approach would be to have a protocol before you ask the question. Before asking AI to complete an important task, for instance, you could ask what information it needs, where the response could go wrong, and what should be checked afterward.
Picture two firms building the same internal tool to help their staffs draft client-facing research summaries. If this were done poorly, your artificial intelligence tool could pull from anything it finds, with no way to trace where an answer came from, and the output would go straight to a client file.
If the tool were built well, on the other hand, the firm could make sure it looks only to approved, vetted sources, that the prompt avoids overconfident language, and that it routes every output to a colleague or supervisor before it reaches a client.
The difference is the discipline wrapped around the tool.
Someone Has To Own It
Here’ s where a lot of firms quietly fall short: Everyone agrees AI needs oversight, but no one is actually assigned to do it.
There doesn’ t have to be a formal committee( especially at smaller advisory shops where there is no one with a dedicated info-security function). It could be that one person is tasked with knowing which tools are in use, which policies apply, and what questions remain unanswered.
What it can’ t be is nobody. If you can’ t name the person who would answer for
how AI was used in a specific client file, that’ s the gap to close first before you add any new tool.
Three Things To Ask Before Trusting The Output
Most firms already have exposure to AI risk, whether their leaders realize it or not. The governance question should not be whether AI is being used, but whether the leadership knows where and how the tech is influencing decisions.
Before accepting any AI-assisted output tied to a client outcome, ask: What information did the model rely on? What would happen if the answer were wrong? What independent source verified the conclusion? If you can’ t answer all three, the output isn’ t ready to leave your desk.
This applies to your vendors as much as your internal technology tools. An advisor’ s responsibility doesn’ t disappear because a third party built the model.
Some questions are worth pressing on before you deploy anything: Where is the client data stored? And is it used to train external models? Can the tool produce documentation regulators would ask for?
That last question isn’ t hypothetical. Your obligation to hold onto communications about investment recommendations doesn’ t disappear just because AI made them. Your firm may be asked during an exam or dispute to produce the AI summary behind the counsel you gave. If that summary isn’ t retained, your firm could discover the gaps at the worst possible time— before it’ s too late to make changes.
Does Your Coverage Reflect How You Actually Work?
You shouldn’ t assume that your existing errors and omissions coverage will automatically extend to AI-assisted work. The terms and exclusions in E & O policies can vary, and courts haven’ t issued a substantial body of rulings on AI liability. You should ask your carrier directly how the policy responds, assess your exposure by use case rather than for the tool, and revisit your coverage as your AI use grows. A claim involving AI-assisted advice can straddle professional liability, cyber coverage, and technology coverage simultaneously, and the gaps tend to live in the seams between those policies, not within any single one.
Closing The Gap
AI doesn’ t create fiduciary responsibility, remove it or share it. It simply changes how decisions get made. Firms that understand that distinction will capture the benefits of AI. Firms that don’ t may not discover the liability gap until they’ re defending a recommendation, complaint or claim, and by then, the question won’ t be whether AI was involved. It will be who was supposed to be watching it.
NICK SINKUS is the chief underwriting officer at Hiscox USA, a leading insurer for more than 600,000 small businesses.
34 | FINANCIAL ADVISOR MAGAZINE | NOVEMBER / DECEMBER 2026 WWW. FA-MAG. COM