Bank AI agents: customers bear risk if hacked
Major banks are tightening terms around third-party AI tools, saying customers remain responsible for transactions and losses linked to automated assistants.

Key takeaways
- Citi updated its Online User Agreement to say customers are responsible for activity through third-party AI agents.
- Citi also says customers may be responsible if an AI tool leaks credentials or is hacked.
- JPMorgan, Bank of America and PNC have made similar changes, while Citizens warned customers by email.
- Anthropic's personal finance tool is read-only and cannot move, manage or invest funds.
In this article
What changed
According to TechCrunch, Citi is the latest major bank to tighten rules around bank AI agents. The lender has updated its Online User Agreement to make clear that customers who allow third-party AI tools to access accounts are responsible for activity carried out through those tools.
The change follows similar policies from JPMorgan and Bank of America last month. PNC, which TechCrunch describes as the eighth largest bank, changed its terms on 18 November 2026. These moves cover AI agents that can be given credentials for bank accounts and other financial portals. Such tools may analyse spending, offer investment ideas, make payments or handle bills. Banks say they will not be held responsible for undoing transactions carried out by automated assistants, whether intentional or unintentional.
What the new terms say
Citi's update goes beyond saying that customers are responsible for their own AI transactions. It also says that if an outside AI tool exposes login details or is compromised, the customer may be responsible for what hackers do with the account. That means a security failure at the AI provider could still leave the customer facing losses.
Citizens sent a similar warning by email this week, in an email reviewed by TechCrunch. The bank told customers they are responsible for actions taken by their AI assistant. Citi told TechCrunch that customers need to know the dangers of handing over bank login details or letting outside tools access accounts. The bank said the update was intended to give clearer information about those dangers.
Why banks are resisting
Banks are uneasy about unreliable AI systems acting inside their networks. They also see personal finance as a competitive threat. TechCrunch reports that AI startups are looking for new revenue, and financial accounts offer a valuable route to fees and customer attention.
The report says markets reacted briefly to the prospect that AI tools could help consumers sidestep needless fees or low-interest deposit accounts. Such accounts can support bank margins. If AI agents become a main way people manage money, banks want to control that channel. Citi's terms say customers can use AI agents provided by Citi without extra liability worries. Consumers may also be cautious about sharing sensitive records with AI tools, making bank and brokerage accounts a new trust challenge.
A wider fight over AI assistants

The banking changes are part of a broader pattern across services. AI labs say their agents will change how people interact with companies, from transport to food delivery. Service providers are now pushing back and setting limits. TechCrunch says these conflicts will shape which AI bets succeed.
Anthropic's approach shows one way to reduce risk. Its new personal finance product, released earlier this month, is described as read-only. It says the tool cannot move, manage or invest funds. That design avoids giving the assistant power to make transactions, but it also limits what the tool can do for customers. The same tension is visible in car-hailing and food service, where AI promises are meeting practical limits.
What it means for account holders
The new terms change the practical risk of using automated financial assistants. A customer who connects an outside AI agent to a bank account may have fewer options if the agent makes a mistake or is compromised. The bank can treat the activity as authorised by the customer.
The policies also create a distinction between outside tools and tools offered by the bank itself. Citi says its own AI agents can be used without additional liability. Read-only products, such as Anthropic's, reduce the chance of unauthorised transactions but offer less functionality. The official terms published by each bank are the controlling guidance for customers.
Sources
Spotted a mistake? See our corrections policy. Read how News by OnDevice works in our editorial policy.



