For years, Michelle Singletary has written about the subtle, enduring ways that people entrust their financial lives to systems that were never intended to safeguard them. Her most recent piece for the Washington Post seems to be an extension of that work, but this time the system in question is one that millions of people actively enter their most private information into, frequently without giving it much thought.
The figures she presents are startling. Approximately 10% of Americans reported using AI to handle their finances a year ago. According to the most recent survey conducted by TD Bank, that percentage has since increased to 55%. It’s even higher among younger adults: 72% of millennials and 77% of Gen Z are now using AI chatbots for financial advice. That behavior is no longer considered niche. Most people are unaware of how quickly this cultural shift is occurring.
This is where it becomes uncomfortable. According to a Cisco survey, 29% of AI users worldwide have input private or sensitive data into chatbots, including financial information, health information, and employment records, despite the fact that 84% of those users expressed concern that their data might be made public. The kind of thing that keeps privacy researchers up at night is the disparity between concern and behavior.
Five categories of information that AI chatbots have no business knowing are listed in Singletary’s column. Social Security numbers, home addresses, employment information that could be used as a weapon in phishing schemes, financial account information, medical records, passwords, and login credentials are among the personally identifiable information on the list. Regardless of how useful the AI appears at the time, she contends that none of these should ever show up in a chat window.
It is difficult to discount the Stanford research she cites. All six of the major AI companies—Amazon, Anthropic, Google, Meta, Microsoft, and OpenAI—use chat data to train their models by default, according to researchers looking at their privacy policies. Some people never forget that information. The researchers cautioned that chatbots are radically altering how people divulge personal information, and that this information may eventually be exploited due to carelessness or malicious intent. That isn’t speculative. It’s a known risk.
The specificity Singletary adds to her column makes it more impactful than a standard tech-privacy warning. She describes two incidents: one in which a clerk at a furniture store attempted to scan her driver’s license into the business’s computer system, and another in which her dentist requested to use her Social Security number as a patient identifier. She pushed back both times. She describes an instinct that applies to AI chatbots just as clearly, if not more so: the question of whether a particular institution truly needs the data it is requesting.
One area that receives insufficient attention in these discussions is employment information. Someone has a convincing detail for a phishing email, a fraudulent loan application, or a fake W-2 request if they know where you work. AI can assist you in negotiating a salary without knowing your employer. Answers to general questions are helpful. Exposure results from specific personal information.
In a separate piece on the subject, Bernard Marr raises an important point: once data enters a public chatbot, there is actually no way to know what happens to it. Most chatbot interfaces simply lack the systems that handle credit card numbers during valid e-commerce transactions, such as encryption, automatic deletion, and layered security protocols. Both authors seem to agree on a helpful mental model: treat anything you enter into a chatbot as potentially public information and act appropriately.
It’s difficult to ignore how much of this boils down to a conflict between caution and convenience. AI chatbots are actually helpful for many things, such as contribution caps, budget-friendly meal planning, and general financial concept explanations. The tool is not the issue. The issue is the increasingly prevalent belief that a chatbot must be reliable enough to store personal data in order to be attentive enough to offer wise counsel. These are very different things, and the difference is more important now than it will likely be until something goes wrong.
