Track · AI in Product3 min read
Ally, Chase and Citi Tell Customers: You Pay for Your AI Agent's Mistakes
Ally, Chase and Citi are warning customers that AI agents count as authorized users — meaning hallucinated payments and mistaken purchases fall on the account holder, not the bank.

Context
Ally Bank emailed customers this month: "you're responsible for any actions that the AI takes."
Target changed its terms in March to treat AI-agent shopping as customer-authorized; Amazon blocked Muse agents in September.
Axios reported in October that major banks detected third-party agents inside customer accounts, some with customer credentials.
JPMorgan CEO Jamie Dimon warned on an October earnings call that outside agents are "too risky" to admit into the company.
Apple's macOS 27.3 beta warns users before granting AI agents system-level computer access.
Ally Bank has told customers in an email this month that they bear full responsibility when an AI agent acting on their behalf goes wrong. "When you use an AI service, you're responsible for any actions that the AI takes," the bank wrote, according to messages reviewed by Business Insider. "AI can make mistakes, hallucinate, misunderstand your instructions, or follow them in ways you didn't intend, which could lead to significant loss."
Ally is not alone. Chase warned customers in a November email that "authorized transactions may not be reversible if your information is misused." Citi placed a banner on its website and mobile app directing users to a caution page stating that "AI assistants can make mistakes, and that means you can be liable for errors it makes." The policy line is consistent across institutions: an AI agent with access to a financial account is treated as an extension of the user, not as an outside intruder.
Why are banks sending these warnings now?
The alerts track the rise of a new class of AI agents that do not merely answer questions but complete tasks online. Earlier models such as OpenAI's ChatGPT and Anthropic's Claude replied with blocks of text. Current agents — Meta's Muse, Instinct AI, and OpenAI's Dots among them — can book flights, reserve restaurant tables, and reorder household goods on a customer's behalf.
That capability has already forced changes in retail. In March, Target amended its customer agreement to state that AI-agent shopping is "considered transactions authorized by you." In September, Amazon blocked Muse agents from making purchases outright, citing safety and reliability concerns.
Where do banks draw the line on access?
Banks are distinguishing between formally authorized account connections to AI tools and customers who simply hand over login credentials. Axios reported in October that major banks had detected third-party agents operating inside customer accounts, in some cases with the customer's permission and credentials supplied.
Every bank message reviewed by Business Insider instructed customers not to share usernames, passwords, or security codes. Wells Fargo took the strictest public stance. "No technology should be trusted to act on a customer's behalf without strong security, clear customer authorization, and safeguards designed to protect their money and financial information," a bank representative said.
JPMorgan Chase CEO Jamie Dimon addressed the same risk on an October earnings call. "If people are using them, they should understand the risk," he said. "In the meantime, we're not going to allow just any outside agent to come in inside the company because that's just too risky for clients and for JPMorgan."
The warning campaign has reached device makers as well. The beta of Apple's latest Mac software, macOS 27.3, displays caution language when a user considers granting an AI agent or application system-level access to the computer.
What does this mean for product teams?
The direction of travel is clear: liability follows authorization. As agents move from answering questions to executing transactions, the institutions that hold the money — and the platforms that host the agents — are formalizing the principle that the human who granted access owns the outcome, including hallucinated purchases and misdirected payments.
via feeds.businessinsider.com (Source)
More from Rebecca Stone
Show full bio
Correspondent covering regulation, policy and their impact on product management at Roadmap File.
72 articles