Major U.S. Banks Shift Fraud and Hacking Liability Onto Customers Using Third-Party AI Agents

Major U.S. retail banks are drawing a hard contractual boundary around autonomous artificial intelligence. This week, Citibank—the third-largest commercial bank in the United States—and eighth-largest PNC Bank updated their online user agreements to warn customers that anyone who connects a third-party AI agent to their bank account bears full financial responsibility if anything goes wrong, including if the AI provider itself is hacked and leaks their login credentials. Citizens Bank issued a parallel notice to customers by email, following similar policy shifts introduced last month by JPMorgan Chase and Bank of America. Taken together, the moves mean the nation’s largest deposit institutions have effectively closed ranks against unapproved "agentic" finance software. Rather than waiting for federal banking regulators to draft new liability standards for autonomous software, banks are rewriting the rules through private contract law—forcing consumers and AI developers alike to confront a stark mismatch between what AI agents promise and who pays when they fail. ### From Credential Sharing to a Liability Trap Under Citibank’s revised Online User Agreement, customers are instructed not to share banking credentials with external AI tools. More consequentially, the bank stipulates that any access, instruction, or transaction carried out through an authorized third-party AI agent is legally treated as if the customer performed it directly. As reported by [TechCrunch](https://techcrunch.com/2026/11/18/more-banks-says-customers-using-ai-agents-are-on-their-own-even-if-they-get-hacked/) and [CryptoBriefing](https://cryptobriefing.com/citibank-customers-liable-ai-agent-risks/), Citi’s language goes a step further than simply disclaiming responsibility for an AI assistant’s miscalculations or accidental bill payments: if the external AI service suffers a data breach or exposes a user’s login details, the customer remains on the hook for subsequent unauthorized withdrawals or fraud carried out by attackers. For consumers accustomed to strong statutory fraud protections on debit and electronic transfers, that clause exposes a critical legal gray area. Under the Electronic Fund Transfer Act and its implementing rule, Regulation E, U.S. banks must generally absorb losses from "unauthorized" electronic fund transfers above a low statutory cap. Yet Regulation E was written around a binary distinction between an account holder and an intruder. When a customer voluntarily hands username-and-password credentials—or broad account-delegation permissions—to an outside software agent, banks argue that subsequent activity falls inside the scope of customer-authorized access, potentially stripping away the fraud-recovery safety net that protects ordinary card and bank users. ### Security Perimeters vs. Protecting Deposit Margins The sudden wave of policy updates across JPMorgan, Bank of America, Citibank, PNC, and Citizens reflects two converging pressures: genuine technical risk and fierce commercial gatekeeping. On the security side, autonomous browser-use agents and personal finance bots frequently rely on raw login credentials or fragile screen-scraping workflows rather than scoped banking APIs. Once an agent holds a full session token, prompt-injection attacks, software bugs, or unprompted tool behavior—a hazard underscored by recent incidents such as [Grok Bot’s unapproved corporate data queries](https://www.mushimu.com/post/9506)—can trigger real-world consequences that automated fraud systems struggle to distinguish from legitimate customer behavior. Speaking on JPMorgan’s third-quarter earnings call, CEO Jamie Dimon framed the danger bluntly, warning against allowing "rogue agents" inside bank infrastructure and stressing that customers using outside tools must understand there is a "liability transfer," as covered by [Quartz](https://qz.com/jpmorgan-blocking-outside-ai-agents-customer-accounts-101326). Yet cybersecurity is only half the equation. Banks also view independent AI agents as a direct competitive threat to their core retail franchises. Financial software capable of continuously scanning balances, spotting hidden fees, and automatically sweeping idle cash into higher-yielding accounts threatens the low-interest deposit spreads that underpin traditional bank profitability. By contrast, Citibank’s updated terms explicitly exempt AI assistants provided directly by the bank itself, while JPMorgan has paired its crackdown on outside agents with internal pilots such as its proprietary cash-forecasting assistant. Controlling the "agentic interface" is rapidly becoming the next battleground over who owns the customer relationship. ### Why Read-Only Open Banking Is Surviving the Crackdown Not all AI finance integrations are being treated equally. When [Anthropic introduced Personal Finance in Claude](https://claude.com/resources/articles/introducing-personal-finance-in-claude) earlier this month for U.S. paid subscribers, the company deliberately engineered around the banks’ red lines. Instead of asking users for bank passwords or automating transfers, Anthropic routed account connections through Plaid on a strictly read-only basis—giving Claude visibility into balances, credit limits, and two years of transaction history while making it technically impossible for the model to move, manage, or invest money. That architectural divide points to where consumer financial AI is likely headed. Credential-sharing agents that ask to impersonate a user inside an online banking portal now carry unacceptable legal exposure for consumers and invite outright technical blocking from institutions. Until open-banking standards evolve to support cryptographically scoped, revocable write permissions—and until regulators clarify whether compromised AI intermediaries qualify as unauthorized fraud under Regulation E—third-party AI tools will likely remain boxed into read-only advisory roles, leaving payment execution firmly behind bank-controlled walls. ### Sources & Further Reading - [TechCrunch (US)](https://techcrunch.com/2026/11/18/more-banks-says-customers-using-ai-agents-are-on-their-own-even-if-they-get-hacked/) - [CryptoBriefing (US)](https://cryptobriefing.com/citibank-customers-liable-ai-agent-risks/) - [Quartz (US)](https://qz.com/jpmorgan-blocking-outside-ai-agents-customer-accounts-101326) - [Anthropic / Claude (Primary Product Documentation)](https://claude.com/resources/articles/introducing-personal-finance-in-claude) ~ArtificialIntelligence ~Banking ~Cybersecurity ~AIAgents ~Regulation

Major U.S. Banks Shift Fraud and Hacking Liability Onto Customers Using Third-Party AI Agents
0
Discuss 0 0
0 replies
Be the first to join this thread.