Citizens Bank, Stanford warn Americans against sharing financial data with AI
Millions of Americans are typing their most sensitive financial details into AI chatbots every day without a second thought.
They enter salary numbers, upload tax documents, and share account balances with tools such as ChatGPT and Gemini, assuming those conversations remain private.
Yet every prompt could become training data, and every detail shared could surface in unexpected ways, from targeted ads based on inferred health conditions to identity theft and financial fraud.
Experts at Citizens Bank and Stanford University are sounding the alarm about how vulnerable consumer data remains in AI environments today.
Citizens Bank executive raises concerns about unregulated AI data sharing
Chris Powell, head of deposits at Citizens Bank, told Money that consumers face real exposure when they hand over personal financial data to AI platforms.
Powell explained that these tools operate outside the regulatory framework governing banks and other traditional financial institutions today.
Financial institutions must follow strict federal and state laws governing how they store, share, and protect customer information and personal data.
AI chatbot companies, by contrast, operate under a patchwork of privacy policies that most consumers never read before agreeing to use them.
In its online safety guidance, Citizens Bank cautions customers against entering personal or key financial information into any AI tool, warning that doing so may put their data at risk.
More AI:
- Micron sits at the center of a red-hot chip rally
- IBM CEO sends blunt message on AI and quantum computing
- Anthropic CEO makes shocking admission about AI
Powell offered a practical workaround for consumers who still want to use AI tools for financial planning and budgeting tasks.
Instead of entering exact salary figures or precise account balances, users can provide approximate ranges to get useful responses from any AI chatbot.
For example, entering a salary band rather than an exact figure still gives a chatbot enough context for tailored advice.
Even ballpark figures yield responses useful for planning, Powell explained to Money.
Stanford researchers reveal AI companies collect user chat data by default
A study from the Stanford Institute for Human-Centered AI found that six major U.S. AI companies use customer conversations to train their language models by default.
The research team, led by Privacy and Data Policy Fellow Jennifer King, examined 28 policy documents across companies, including OpenAI, Google, Meta, Anthropic, Microsoft, and Amazon.
King warned that sensitive information shared in conversations with ChatGPT, Gemini, or other leading models may be collected and used to train these models.
That includes data contained in uploaded files, not just text typed directly into the prompt window, according to Stanford.
We have hundreds of millions of people interacting with AI chatbots, which are collecting personal data for training, and almost no research has been conducted to examine the privacy practices for these emerging tools.
The Stanford researchers highlighted a scenario showing how harmless prompts can cascade into real consequences for everyday consumers using these tools.
Asking a chatbot for heart-friendly recipes could lead the algorithm to classify you as health-vulnerable, a determination that could eventually reach insurance companies.
King concluded that society needs to weigh whether gains in AI capabilities from training on chat data justify the considerable loss of consumer privacy.
The Stanford team recommended comprehensive federal privacy regulation and affirmative opt-in requirements before user data is fed into any model training systems.
Security research quantifies how much sensitive data Americans share with AI
Cybersecurity firm Harmonic Security has measured just how widespread the oversharing problem has become among employees using AI tools at work, findings that underscore the broader risks consumers face when sharing sensitive data with AI platforms.
The company found that 4.37% of all AI prompts contained sensitive information, and more than 20% of file uploads included sensitive data, Money reported.
The safest approach right now is to avoid putting any confidential financial information into a large language model, said Harmonic Security CEO Alastair Paterson in an email to Money.
Paterson noted that oversharing has become a persistent pattern, with many users uploading employer data to write financial reports without considering the downstream risks.
Ramayya Krishnan, a professor of management science and information systems at Carnegie Mellon University, added that standard versions of ChatGPT and Gemini store conversation history that could be at risk if a user's AI account is compromised.
He also told Money that a subset of user conversations is sampled and reviewed by OpenAI and Google employees to improve quality.
In a worst-case scenario, someone with access to those stored conversations could steal an identity or commit financial fraud using data a consumer casually shared.
That risk makes the combined warnings from Citizens Bank, Stanford, and cybersecurity researchers difficult to dismiss for any American using AI tools today.
Financial details Americans should never enter into AI chatbot prompts
Experts interviewed by Money identified specific categories of data that consumers should treat as off-limits in any AI conversation or file upload today.
Citizens Bank's AI safety guidance echoes this advice, warning customers against entering personal or sensitive financial information into AI tools due to the risk of exposure.
Financial data categories to keep out of AI tools
- Bank account information, including routing and account numbers, which could enable unauthorized transfers and withdrawals
- Social Security numbers, investment account details, and brokerage login credentials that could expose retirement savings to bad actors
- Passwords, exact wage and paycheck information, transaction details, account balances, and sensitive tax documents
Source: Money's piece, titled Why You Should Never Share Your Financial Data With ChatGPT
The AI chatbot privacy gap won't close itself
AI chatbots have become default tools for millions of Americans managing their money, but the infrastructure protecting that data hasn't kept pace with adoption.
Citizens Bank, Stanford, and cybersecurity researchers interviewed by Money all point to the same disconnect: Consumers treat these platforms as private financial advisors, while the companies behind them, according to the Stanford study, treat user input as training material by default.
The Stanford team called for comprehensive federal privacy regulation and affirmative opt-in requirements before user data is used to train models.
Until those guardrails exist, the experts quoted by Money say the risk of sharing sensitive financial information with an AI chatbot sits with the individual user, not the platform.
Related: AI is getting worse as Google and Anthropic nerf AI models and limit usage
The Arena Media Brands, LLC THESTREET is a registered trademark of TheStreet, Inc.
This story was originally published May 29, 2026 at 9:47 AM.