First Internet Bank has opened a permission-controlled channel between customer accounts and two of the most widely used generative artificial intelligence platforms, allowing eligible business and personal banking customers to ask ChatGPT or Claude questions based on their own financial activity.

The service creates a read-only connection between supported versions of OpenAI’s ChatGPT, Anthropic’s Claude and qualifying First Internet Bank accounts. Customers can authorize the assistants to review selected account information and then use ordinary language to request analysis of balances, transactions, spending trends, savings activity or business cash flow.

The launch represents a shift from using general-purpose AI for hypothetical financial questions toward allowing it to work with verified, customer-specific information. Without account access, an AI assistant can offer generic budgeting frameworks or explain financial concepts, but it cannot determine how much a particular customer spent, which vendors received the most money or whether recent cash inflows are keeping pace with expenses.

First Internet Bank’s connection is intended to close that information gap. A personal customer could ask how dining expenditure changed from one month to the next, identify subscriptions that appear repeatedly in posted transactions or calculate savings accumulated during the year. A small-business customer could review recent cash-flow movements, examine vendor spending or assess whether available funds appear sufficient for upcoming payroll obligations.

The capability is powered by Model Context Protocol, commonly known as MCP. The protocol provides a standardized structure through which an AI application can connect to an external data source or service. In this case, the bank exposes approved information through a controlled connection rather than requiring customers to download statements, copy transaction records into prompts or manually assemble spreadsheets.

MCP is becoming an important infrastructure layer in the developing market for AI-connected financial services. It can reduce the need to build a separate, proprietary integration for every combination of bank, data source and AI assistant. A financial institution can instead provide a standardized connection that supported AI platforms can recognize, subject to the institution’s authentication, authorization and governance controls.

For First Internet Bank customers, the connection is optional. Accounts are not linked automatically, and customers decide whether to establish access. The bank says customers remain able to review, alter or revoke their permissions, an important distinction as financial institutions experiment with distributing services through applications they do not directly operate.

The initial implementation is deliberately limited to information retrieval and analysis. ChatGPT and Claude cannot use the connection to transfer money, pay bills, initiate automated clearing house transactions, change account settings or otherwise execute financial activity. The assistants can interpret authorized information, but they do not receive transactional authority.

That boundary separates the service from more expansive forms of agentic banking in which software agents may be permitted to initiate or manage financial actions within predefined limits. First Internet Bank is beginning with a lower-risk model focused on visibility and analysis, keeping payment execution inside the bank’s established digital channels and approval systems.

The bank also says customer online-banking credentials are not shared with ChatGPT or Claude. Customers authenticate through the connection process, while the AI assistant receives only the information made available through the authorized interface. According to First Internet Bank, customer banking data accessed through the capability is not used to train the underlying AI models.

Those protections are central to the product’s commercial viability. Consumers and companies may be willing to use conversational tools to review financial information only if they can determine which accounts are connected, understand the scope of access and confidently terminate the connection. Clear distinctions between account data, authentication credentials and model-training data will therefore be important to adoption.

The supported account set includes business and personal checking, savings and money market accounts, according to the bank’s announcement. First Internet Bank’s customer guidance also indicates that the service is designed primarily around balances, posted transactions and related account information rather than unrestricted access to every document held in online banking.

Pending transactions are not currently returned through the MCP connection, according to the bank’s technical support information. The system may provide metadata about statements or tax documents, but it does not download and display the complete documents through the assistant. Customers must continue using online banking for records that fall outside the connection’s available tools.

Such limits matter because conversational answers can appear comprehensive even when the underlying data set is incomplete. A cash-flow question based only on posted transactions may not reflect pending debits, recently initiated payments or obligations held outside First Internet Bank. Customers will therefore need to understand the boundaries of the connected data before relying on an answer for an immediate financial decision.

The bank’s materials similarly present the service as a way to improve access to information rather than as a replacement for financial judgment. An assistant may organize transactions and identify patterns more quickly than a customer working through account statements, but its response depends on the data available, the wording of the question and the model’s ability to interpret the records correctly.

Availability is also tied to paid versions of the two AI platforms. First Internet Bank said the connection is available for Claude Pro, Max, Team and Enterprise plans, as well as ChatGPT Pro, Business and Enterprise plans. That positioning places the service initially with customers already paying for advanced AI functionality rather than offering it universally through the bank’s own application.

The arrangement gives First Internet Bank a presence inside platforms where customers may already conduct research, write documents, analyze business information or plan household finances. Instead of requiring a user to move between a banking dashboard and an AI tool, the connection brings authorized banking context into the conversational environment.

For small businesses, that distribution model could be particularly relevant. Many smaller companies do not employ a full financial planning and analysis team and may rely on owners, bookkeepers or outside accountants to assemble periodic cash-flow reviews. Natural-language access to account activity can shorten the process of answering routine operational questions, even if more consequential decisions still require formal accounting records and professional review.

David Becker, chairman and chief executive officer of First Internet Bank, said the service was intended to extend forms of financial analysis that historically required spreadsheets, manual reviews or specialized planning resources. The bank’s stated objective is to let consumers and businesses derive useful answers from information already held in their accounts without adopting another dedicated analytical system.

First Internet Bank is a natural participant in the development of an AI-based banking channel. The institution opened in 1999 as a branchless bank and built its operating model around remote delivery rather than a traditional retail branch network. It reported $5.7 billion in assets as of March 31, 2026 and provides deposits, consumer lending, small-business banking, specialty finance, commercial lending and treasury-management services.

The implementation also highlights the growing role of digital-banking infrastructure providers in determining how quickly smaller and midsized institutions can enter the AI market. First Internet Bank’s support materials reference a Narmi MCP connector, while Narmi has introduced a suite of applications designed to bring bank account data and financial tools into ChatGPT and Claude through secure, read-only connections.

Narmi’s broader MCP offering includes functions for reviewing cash-flow history, modeling budgets and presenting bank product information. The infrastructure is designed to let financial institutions control which capabilities are enabled while maintaining their branding and governance policies. That approach could allow community banks and credit unions to introduce conversational services without independently constructing every component of the underlying integration.

For banking software providers, MCP creates a potential new product category alongside mobile banking, account opening, payments and data aggregation. Banks may increasingly evaluate technology vendors not only on the quality of their own web and mobile interfaces but also on their ability to distribute authorized services through external AI hosts.

That development could alter the competitive significance of the bank’s primary application. Digital banking has historically centered on persuading customers to spend more time inside an institution-controlled website or mobile app. Conversational banking instead assumes that some interactions will occur inside a third-party assistant, with the bank supplying trusted data and regulated infrastructure in the background.

The model presents strategic opportunities but also governance challenges. Banks must decide which information an assistant can retrieve, how consent is recorded, how sessions expire, how access is revoked and how customers are informed when an answer does not include pending or externally held activity. They must also monitor whether third-party platforms change their technical requirements, privacy policies or supported subscription tiers.

MCP itself is an interoperability standard rather than a complete banking-security framework. Its specification emphasizes explicit user consent, access controls and transparency, but financial institutions and their technology partners remain responsible for implementing authentication, data protection, auditing and capability restrictions around each connection.

Read-only access reduces the consequences of an incorrect response or malicious instruction because the assistant cannot directly execute a transfer. It does not eliminate every risk, however. Sensitive transaction information can reveal business relationships, household behavior, income patterns and recurring obligations. Customers must therefore treat access to conversational financial data with the same care applied to other applications connected to a bank account.

Accuracy will be another focus. Large language models are designed to produce natural-language responses, and the clarity of their presentation can make an answer appear more definitive than the underlying calculation warrants. Banks offering such integrations will need to help customers distinguish between retrieved facts, model-generated summaries and forward-looking interpretations.

The First Internet Bank rollout stops well before automated financial decision-making. It does not authorize the assistant to rebalance funds, select financial products, approve invoices or schedule payments. Nevertheless, the integration establishes technical and behavioral foundations that could support more advanced functions if banks, regulators and customers become comfortable with tightly governed AI access.

Future development across the industry may include approval-based workflows in which an assistant prepares a transaction but a customer authorizes it through a separate banking screen. Other possibilities include treasury alerts, invoice analysis, personalized product discovery, automated reconciliation and connections spanning multiple financial institutions. Each additional capability would increase convenience while also raising the importance of identity controls and auditability.

For now, First Internet Bank is positioning conversational AI as an analytical layer above conventional banking rather than a replacement for it. The account remains at the regulated institution, the bank retains control over the connection, and financial actions stay within established payment and digital-banking systems.

The commercial test will be whether customers find the conversational interface materially faster and more useful than existing transaction-search, budgeting and reporting tools. Adoption will also depend on whether users are comfortable granting an external AI platform access to account information, even when the connection is optional, revocable and restricted to reading data.

If the model gains traction, it could give banks a new way to remain relevant as AI assistants become a common starting point for personal and business decisions. Rather than allowing those assistants to respond with incomplete assumptions, institutions can provide permissioned access to verified information while preserving boundaries around execution.

First Internet Bank’s launch therefore marks an early practical test of banking through an AI interface. Its immediate functions are straightforward—searching activity, summarizing patterns and answering questions—but the underlying change is broader. The bank is treating ChatGPT and Claude as potential customer-access channels, with consent-controlled financial data serving as the bridge between conversational software and regulated banking infrastructure.