Audrey Woods, MIT CSAIL Alliances | June 22, 2026
Have you asked an AI model for financial advice? You’re not alone. In the last year, the number of Americans who reported using AI to help manage their personal finances went from 10% to 55%, according to TD Bank. But how useful is that advice? Can it be trusted? What happens when broad swaths of the public rely on commercial LLMs to guide their investment decisions?
Andrew W. Lo, MIT Professor in the Sloan School of Management, CSAIL PI, and Faculty Co-Director of FinTechAI@CSAIL, says, "The question is: `Do large language models have our back?’ The answer is no, not yet.”
MAJOR HURDLES: CONFIDENCE, HALLUCINATIONS, AND RESPONSIBILITY
Whatever the experts say, millions of people are already turning to AI for personal financial advice. For Professor Lo, that isn’t necessarily a bad thing. Current commercial models like ChatGPT 5.2 can provide fairly solid guidance for general questions or broad overviews of topics like why it’s important to diversify investments. But one of the critical issues in using AI for financial advice is the way that advice is delivered. Professor Lo says, "No matter what you ask, it'll always come back with an answer that sounds authoritative, even if it's not.” AI is notoriously bad at admitting when it is unsure, or hedging if it doesn’t have all the information, which means that it will inevitably “communicate both good and bad financial advice with the same pleasant and convincing affect.”
Another reason LLM financial advisors cannot replace human financial advisors—yet—is the risk of hallucinations. With any AI model there is always the risk of it inventing an entirely fabricated response or source, and if the answer is given in that self-confident tone a user might never suspect the information is wrong. Professor Lo acknowledges hallucination is not unique to AI models and human advisors can also be very wrong, which is why it’s always worth weighing personal financial advice and making a well-informed decision. But when it comes to users delegating financial decisions to AI agents or allowing them to, for example, make independent investments, Professor Lo says, “That would be a bad mistake at this time.”
Ultimately, AI cannot be a qualified and trustworthy financial advisor until the issue of legal responsibility is worked out. Professor Lo is actively studying the question of fiduciary duty, or the legal standard that requires an advisor to place a client's interests above their own. Human advisors who breach that duty face regulatory penalties, civil liability, and in some cases criminal exposure. AI systems currently do not. As Professor Lo put it to CNBC, if they make a mistake, AI models "don't have the ability to suffer consequences to the same degree that a human advisor does." Without that accountability, the principle of acting in the client's best interest "has no teeth."
Professor Lo's group is investigating how a software system can be designed to satisfy the fiduciary standard. Any model deployed by a brokerage firm could be instructed to favor products that generate commissions, and large commercial models like Claude might have their own constraints that conflict with what’s best for a given individual. Professor Lo’s goal is to develop a model that is a true fiduciary, always prioritizing what’s best for the client and tailoring advice to that individual’s needs, both practical and emotional. To train such a model, he plans to feed it all US laws, regulations, and court cases involving questions of financial ethics, which he describes as “a fossil record of all the ways that bad actors have attempted to exploit unsuspecting retail and institutional clients." Ideally, this will teach the AI what not to do. Combined with specialized modules to produce human-like empathy, humility, and fairness, he wants to create something robust and trustworthy for everyday consumers. Professor Lo plans to release this model free to the public.
THE FUTURE OF AI FINANCIAL ADVISORS
Professor Lo does not believe AI will replace human advisors so much as augment them and expand their capabilities. His expectation is that roughly 80% of advisory work—routine questions, account maintenance, and standard planning—can be supported or handled by AI, freeing human advisors to concentrate on the remaining 20% that requires human judgment, creativity, and personal trust. Complex tax situations, liquidity events, and decisions intertwined with family or health considerations will still require a “human touch” for the foreseeable future.
For individuals using current AI models for financial guidance, Professor Lo recommends leveraging these systems for orientation rather than execution: framing questions, mapping options, and preparing to talk with human advisors rather than acting directly on the model's output. "[AI] can be very useful in providing different options and in describing how those options might work, but you should always remember that the advice that they give you could be wrong." It’s especially important to think deeply about singular or idiosyncratic advice. "When it comes to very, very specific calculations regarding your own personal financial situation, that's where you have to be very, very careful.”
Will AI ever get to the point where it’s making independent decisions about our investments? It will require better models, policy changes, government protections, and an improvement in consumer understanding. But Professor Lo says, “I do believe that will eventually happen.”
To learn more about Professor Lo, listen to his CSAIL Alliances Podcast Episode or read his CSAIL Alliances Spotlight.
