Privacy & Finance
ChatGPT Can Now See Your Bank Account. The Feature Is Good. The Timing Is Terrible.
On May 13, 2026, a federal class action was filed in California accusing OpenAI of secretly routing ChatGPT users' private conversation data to Meta and Google without consent. On May 15 — two days later — OpenAI launched a feature that lets ChatGPT connect directly to your bank accounts, credit cards, and investment portfolios. The feature is genuinely well-built. That's not the problem.
What ChatGPT finance actually does
Through a partnership with Plaid — the same infrastructure that powers Venmo, Robinhood, and 8,000 other apps — ChatGPT Pro subscribers in the US can now link accounts from more than 12,000 financial institutions, including Chase, Schwab, Fidelity, Robinhood, and American Express. Once connected, ChatGPT generates a live dashboard of your spending patterns, portfolio performance, active subscriptions, and upcoming bills.
The more interesting part is the conversation layer. Unlike a budgeting app that shows you a pie chart and makes you figure out what it means, ChatGPT lets you ask things like "what are my three biggest discretionary expenses this month?" or "if I cut dining to $200, how much faster do I pay off my credit card?" These are questions Mint, YNAB, and Copilot Money technically have the data to answer but have never been able to respond to conversationally. You've always had to navigate menus and synthesize the answer yourself. ChatGPT just answers — using GPT-5.5 for financial reasoning, with a dedicated "Financial memories" system that persists context across sessions.
What it doesn't do (and why that matters)
To OpenAI's credit, the architecture is well-scoped. The Plaid connection is read-only — ChatGPT can see your balances and transactions but cannot move money, trade stocks, or pay bills. Your bank credentials never touch OpenAI's servers; the authentication handshake happens entirely inside Plaid's OAuth flow. Account numbers are not exposed — only enriched transaction data. And if you disconnect, OpenAI says it deletes synced data within 30 days.
These are real safeguards. If any other fintech startup launched with this architecture, the security community would call it the right way to do AI-powered personal finance. Plaid is battle-tested. Read-only is table stakes. The opt-out for model training exists.
But OpenAI is not any other fintech startup.
The lawsuit the launch was hiding behind
The class action, Couture v. OpenAI Global, LLC (Case No. 3:26-cv-03000), was filed May 13 in the Southern District of California. The complaint alleges that OpenAI embedded Meta's Facebook Pixel and Google Analytics tracking code inside the ChatGPT website, causing users' query topics, account identifiers, and email addresses to be silently transmitted to Meta and Google. The suit claims this violated the federal Electronic Communications Privacy Act and California's Invasion of Privacy Act.
A separate, nearly identical complaint had been filed May 6 in the Northern District of California. OpenAI did not respond publicly to either suit before launching the bank-account feature.
Let that sink in. The company asking you to hand over your Chase login and Schwab portfolio is currently defending itself against allegations that it was secretly beaming your ChatGPT conversation topics to Facebook and Google. Not a theoretical privacy risk. Not a "could happen someday." An active federal case about the exact kind of data handling you would need to trust to connect your finances.
The trust paradox
Here's the uncomfortable truth: the feature itself is not the problem. It is the gap between the feature's technical quality and the institution's trustworthiness that makes this moment so strange.
The product team clearly thought about boundaries. Read-only access. Plaid-handled auth. Thirty-day deletion. Separate memory namespace. These are the decisions of people who understand what they're asking users to share and want to do it responsibly.
But a well-designed feature shipped by a company under active litigation for mishandling user data is not a product launch — it's a stress test. And the stress is landing on the user: trust our architecture, not our track record.
I have been using ChatGPT Pro since it launched. I find it genuinely useful. When the Finances option appeared in my sidebar last week, I hovered over it for about three seconds and closed the tab. Not because I think Plaid is insecure. Not because I think the read-only scope is a lie. Because I cannot reconcile "we never shared your data" with "actually there's a lawsuit about that, and we haven't addressed it." That gap — between what the feature promises and what the company has earned the right to ask for — is the entire story.
OpenAI wasn't even first
The narrative that OpenAI is "boldly going where no AI assistant has gone before" is marketing, not fact. Perplexity shipped bank, credit card, and loan support via Plaid on April 9, 2026. Anthropic's Claude got a personal finance connector through the Era MCP on May 6. ChatGPT is late to its own party.
What makes this launch different from Perplexity and Claude is the institutional context. Neither Perplexity nor Anthropic is currently defending a federal privacy lawsuit. Neither has an active class action alleging they secretly routed user data to ad platforms. The technical integration is table stakes at this point — the differentiation is trust, and OpenAI is launching from a deficit.
The Mint refugees are the real target
Mint shut down in early 2024, scattering 3.6 million users across alternatives like YNAB ($14/month), Monarch Money ($15/month), and Simplifi by Quicken ($95/year). Those apps are solid but were all built before large language models existed. They show you charts. They don't answer questions.
ChatGPT Pro costs $200/month — ten times what most budgeting apps charge. But OpenAI's pitch is not "switch your budget app." It's "you're already paying $200 for coding help, research, and image generation — here's personal finance included." For the Venn diagram of "ChatGPT Pro subscriber" and "person who would benefit from AI-powered money advice," the overlap is a circle. OpenAI knows this. The company says 200 million users already come to ChatGPT monthly with money questions. This feature is not a new market — it's a better answer to questions people are already asking.
What you should actually do
If you're on ChatGPT Pro and curious about the finance feature, here's the honest assessment:
The architecture is sound. Plaid, read-only, no full account numbers, 30-day deletion — these are the right decisions. If the feature shipped under a different company's logo, I'd recommend it without hesitation.
The company context is not. OpenAI has not addressed the Couture lawsuit publicly. Until it does — with specifics, not a blog post about "our commitment to privacy" — connecting your financial accounts means betting that the plaintiff's allegations are wrong. That's a bet you have to make for yourself, but you should make it with your eyes open.
If you do connect: Opt out of model training first (Settings → Data controls). Use private chats — OpenAI says finance data won't be used in shared conversations, but private mode is a cleaner boundary. Review what Plaid is sharing during the connection flow rather than clicking through. And know that you can disconnect anytime and trigger the 30-day deletion window.
If you don't: The competitive landscape is filling in fast. Claude's personal finance connector exists. Perplexity ships updates weekly. Six months from now, AI-powered finance will be table stakes across every major assistant. Waiting costs you nothing except the convenience of having it all in one chat interface.
The feature is good. The timing is terrible. And the real question is not whether the product works — it's whether you trust the company behind it enough to hand over the one dataset that reveals more about your life than any other.