OpenAI seeks bank‑account access

- OpenAI’s latest consumer-finance move is not just a new feature. It is a decision to let ChatGPT ingest regulated, high-sensitivity financial data through Plaid, beginning with a U.S. preview for Pro subscribers announced on May 15. (techcrunch.com) - The mechanics matter. OpenAI said Pro users in the U.S. can connect accounts through Plaid and pull in data from more than 12,000 financial institutions, then use ChatGPT to review spending, subscriptions, portfolio performance and upcoming payments. (techcrunch.com) - That pushes ChatGPT into a different trust category. A chatbot summarizing a PDF is one thing; a chatbot reading checking-account flows, brokerage balances and card activity is another. American Banker framed the issue as whether users should trust OpenAI with bank-account data, and the product design itself shows why that question is now practical rather than theoretical. (americanbanker.com) - OpenAI’s setup, as described by TechCrunch, routes account linking through Plaid rather than asking users to hand credentials directly to ChatGPT. Users can connect accounts from firms including Schwab, Fidelity, Chase, Robinhood, American Express and Capital One, and OpenAI said users can later remove connections in Settings. (techcrunch.com) - OpenAI also said synced financial data is removed from ChatGPT within 30 days after a user disconnects a service, and users can view and delete “financial memories” from the Finances page. Those details are important because they define the product’s retention and user-control promises at launch. (techcrunch.com) - The unresolved issue is not whether Plaid can connect accounts. It is who bears responsibility if an AI system misreads the data, gives flawed guidance, or encourages a harmful action. The American Banker commentary raises that governance and liability question directly, and neither the Plaid connection itself nor the product demo answers it on its own. (americanbanker.com) - OpenAI is making this trust expansion while also refining how it charges for heavier usage inside ChatGPT. In a Help Center update posted today, OpenAI said users on ChatGPT plans can purchase credits after hitting included limits instead of upgrading plans immediately. (help.openai.com) - Those credits are a pay-as-you-go add-on, not a replacement for the base subscription. OpenAI said included usage is consumed first, after which eligible usage draws from the purchased credit balance. (help.openai.com) - The current scope is narrower than “buy credits for anything.” OpenAI said credits can currently be used with Codex for Plus and Pro users and with ChatGPT for Excel, with flexible pricing for the Excel product taking effect on May 5, 2026. Free and Go users who hit Codex limits are prompted to upgrade to Plus instead of adding credits. (help.openai.com) - OpenAI also added auto top-up. The Help Center says eligible Plus and Pro users can set a minimum and target balance, after which OpenAI automatically purchases only enough credits to restore the balance using the default payment method on file. (help.openai.com) - Separately, OpenAI has published a more explicit Codex pricing framework across workplace and institutional plans. The Codex rate card says the company now lists token-based credit rates for Plus, Pro, Business, Enterprise, Edu, Health and Gov plans, and that pricing was updated on April 2, 2026 to align with API token usage rather than per-message pricing, with the change extended on April 23 to all existing Enterprise plans, including Edu, Health, Gov and ChatGPT for Teachers. (help.openai.com) - The rate card gives concrete examples: GPT-5.5 is listed at 125 credits per 1 million input tokens, 12.50 credits per 1 million cached input tokens and 750 credits per 1 million output tokens. OpenAI also says Codex costs about $100 to $200 per developer per month on average, with wide variation depending on model choice, automations, instance count and use of fast mode. (help.openai.com) - Taken together, the two changes show OpenAI doing two things at once: moving ChatGPT into more sensitive personal workflows and making usage beyond bundled limits easier to monetize. That is not speculation; it is the combined effect of the Plaid-based finance product on one side and the new credit-and-rate-card structure on the other. (techcrunch.com) - The practical consequence is that consent, retention, deletion, auditability and responsibility now matter more than they did when ChatGPT was mostly a text box for general questions. Once a product can read live financial-account data and charge incrementally for advanced agentic work, the operational burden moves beyond model quality to governance. That inference follows from the product details OpenAI and outside reports have already published. (techcrunch.com) - The next things to watch are concrete. OpenAI said the personal-finance tools are in preview for ChatGPT Pro users in the U.S. on web and iOS, and it said Intuit support is planned. On pricing, the live Help Center pages are now the clearest record of where credits apply and how Codex usage is metered across plan types. (techcrunch.com)