OpenAI adds bank‑linked finance tools

OpenAI’s new finance feature is a product launch, but it is also a test of how far the company is willing to let ChatGPT operate inside a high-stakes category. On May 15, OpenAI said it had begun rolling out a preview of a personal finance experience in ChatGPT for Pro users in the United States. The feature lets eligible users connect financial accounts through Plaid, view a dashboard of spending and portfolio activity, and ask questions based on their own financial data. OpenAI said the rollout starts on web and iOS and supports more than 12,000 financial institutions. (openai.com) The immediate product shape is straightforward. Connected users can see spending, bills, subscriptions, upcoming payments, net worth and investment information in one place, then ask ChatGPT to help with budgeting, debt payoff, savings goals, major purchases and portfolio review. OpenAI’s help documentation says the tool can explain tradeoffs and organize information, including for tax-related review, but it does not file taxes. (help.openai.com) A few details matter because they show what OpenAI is trying to ship first. OpenAI said the launch is a preview for a smaller group of Pro users in the U.S., not a full public release, and that it plans to learn from early use before expanding to Plus users and eventually wider availability. The company said users can start from a new Finances page in the ChatGPT sidebar or by invoking “@Finances” in chat. It also said Intuit support is coming soon. (openai.com) Plaid is the key infrastructure layer. OpenAI is using Plaid to connect bank, credit card, brokerage and other financial accounts, and both companies say the integration is meant to ground ChatGPT’s answers in live account data rather than generic personal finance advice. Plaid said its network covers more than 12,000 institutions and that its transaction models help classify messy bank data into something a product like ChatGPT can use. (openai.com) The limits are just as important as the features. OpenAI says ChatGPT is not a fiduciary, registered investment adviser, broker-dealer, tax preparer or law firm. The help page says responses are for informational and planning purposes only, and that users remain responsible for decisions with legal, tax or investment consequences. OpenAI also says ChatGPT cannot move money, pay bills, make trades, change retirement contributions, open or close accounts, or alter account settings. (help.openai.com) That framing reflects a broader safety problem for OpenAI, not just a product disclaimer. Financial questions are often personalized, consequential and cumulative: a model may not be asked “Should I buy this stock?” in isolation, but instead after a long thread about debt, cash flow, savings and risk tolerance. One day before the finance launch, OpenAI published a separate safety update saying ChatGPT had been improved to recognize when risk emerges over time by identifying subtle or evolving cues across a conversation and using that context to respond more carefully. OpenAI described that work in the context of sensitive conversations generally, rather than finance specifically, but the timing shows the company is building more context-aware safeguards as it pushes ChatGPT into domains where mistakes carry higher stakes. That is an inference based on the sequence of OpenAI’s own announcements. (openai.com) The commercial logic is also visible. OpenAI says more than 200 million people already come to ChatGPT each month for budgeting, investment questions and goal planning. Giving those users a way to connect actual accounts turns that behavior into a stickier product, at least for paying subscribers. For now, the feature is limited to ChatGPT Pro in the U.S.; OpenAI’s help center says Pro plans start at $100 a month, with a higher-usage $200 tier also available. (openai.com) What comes next is unusually explicit. OpenAI says it is rolling the product out gradually, plans to expand beyond the initial preview group, and intends to bring the experience to Plus users after learning from early usage. (openai.com)