
Major US banks are tightening their rules around third-party Artificial Intelligence agents, warning customers that they may be responsible for transactions and other account activity carried out by AI tools they authorise. JPMorgan Chase, Bank of America, Citibank, Citizens and PNC have reportedly updated or reinforced their terms to clarify that customers who give external AI agents access to their Banking accounts may bear the consequences if something goes wrong.
The changes come as AI assistants become increasingly capable of interacting with financial services. Instead of simply answering questions about spending habits, modern agents can connect to online accounts, analyse transactions, navigate websites and potentially initiate purchases or other actions. That convenience introduces a difficult question: who is responsible when an AI agent makes an unauthorised-looking transaction, misunderstands an instruction or falls into the hands of a hacker?
The banks’ position is becoming clearer. When customers voluntarily provide account access to an independent third-party AI agent, the resulting activity may be treated as authorised by the customer. That could leave account holders facing disputes over transactions they did not personally initiate or fully understand.
At the same time, major AI companies are exploring more restricted ways to connect assistants with financial information. Google’s Gemini and Anthropic’s tools are among the examples of services moving towards read-only access, allowing users to analyse financial data without necessarily giving an AI agent the ability to spend money.
Why Major US Banks Are Changing Their AI Agent Rules
Traditional online banking generally assumes that customers control their own login credentials and personally authorise financial activity. A customer signs in, reviews a transaction and confirms the action through the bank’s website or application.
AI agents complicate that arrangement. Some tools can interact with websites and connected services on a user’s behalf, potentially carrying out multiple steps without requiring the customer to click through every screen. Depending on the permissions granted, an agent may be able to access financial information or perform actions that affect an account.
For banks, this creates a liability problem. If a customer gives an external AI service their banking credentials or authorises it to access an account, the bank may have difficulty distinguishing between an action deliberately requested by the customer and one generated by an AI system that misunderstood an instruction.
The updated terms are intended to clarify where responsibility sits. Rather than accepting automatic liability for every action taken by an independent AI agent, banks are warning customers that granting third-party access can make them responsible for activity conducted through that service.
Citibank Warns Customers They Are Responsible for Third-Party AI Activity
Citibank has joined JPMorgan Chase and Bank of America in addressing the risks associated with customers allowing external AI tools to interact with their accounts. Citizens and PNC have reportedly made similar changes to their customer agreements.
A Citibank spokesperson explained that the bank wanted customers to understand the risks of sharing banking credentials or authorising third-party AI agents to access their accounts.
“We recently updated our Online User Agreement to provide greater clarity about those risks and to reinforce that customers who choose to grant account access to third-party AI agents are responsible for activities conducted through those agents.”
The statement reflects a broader effort by banks to establish clearer boundaries between their own digital services and external AI platforms.
Citibank’s position also makes an important distinction between third-party tools and AI features provided directly by the bank. The warning concerns independent services that customers choose to connect to their accounts. It should not automatically be interpreted as meaning that customers face the same additional liability when using AI tools supplied and controlled by their bank.
What Does This Mean for Citibank Customers?
If a customer gives an external AI agent access to a bank account, the bank may treat activity performed through that authorised connection as activity approved by the customer. That could create complications if the agent misunderstands a request, performs an unintended action or is compromised.
For example, imagine a user asking an AI assistant to review monthly spending and identify unnecessary subscriptions. If the agent has broader account permissions than necessary and accidentally initiates a payment or changes an account setting, the customer could face a dispute about whether the activity was authorised.
The risk becomes more serious if an attacker gains control of the AI account or exploits the agent’s access. A compromised agent may be able to interact with connected services using permissions that the customer previously granted.
However, the updated terms do not mean every disputed transaction automatically becomes the customer’s responsibility under every circumstance. Applicable consumer-protection laws, payment regulations, the bank’s specific agreement and the facts surrounding an incident can still matter. Customers should not assume that a contractual warning eliminates all rights to report or challenge fraudulent activity.
Why Giving an AI Agent Your Banking Password Is Risky
One of the most basic Cybersecurity principles is to avoid sharing login credentials. Banking passwords, authentication codes and other security details are intended to protect accounts from unauthorised access.
AI agents introduce a new complication because some services need access to connected accounts to perform useful tasks. A user might want an assistant to examine spending, identify recurring payments, compare bills or prepare a purchase. Giving it access can make those activities easier, but it also creates another point of potential exposure.
There are several ways this can go wrong:
- Incorrect instructions: An AI agent may misunderstand a request and take an action the user did not intend.
- Excessive permissions: A tool given broad account access may be capable of doing more than the user originally expected.
- Account compromise: If an attacker gains access to the AI account, connected services or authentication sessions, they may be able to misuse existing permissions.
- Website manipulation: An agent navigating external websites could encounter misleading instructions, malicious content or fraudulent payment pages.
- Transaction mistakes: An agent may select the wrong amount, recipient, product or payment option while completing a multi-step task.
- Limited accountability: Banks and AI providers may disagree over which party is responsible when an automated action causes financial damage.
The problem is not simply whether AI can make mistakes. Human users make mistakes too. The difference is that an autonomous agent can carry out several actions quickly, sometimes across multiple services, before the account holder notices something has gone wrong.
Google Gemini Introduces a Read-Only Approach to Financial Data
One response to these concerns is to limit what an AI assistant can do with financial information. Google’s Gemini has introduced a read-only approach through Google Wallet that allows users to analyse spending and track payments without giving the agent the same purchasing authority that a fully connected financial account could provide.
This model separates access to information from permission to move money. An AI assistant may be able to examine available transaction data, identify spending patterns and answer questions about payments, while being prevented from independently initiating purchases through that connection.
For example, a user could ask Gemini to identify their largest spending categories, find recurring payments or summarise recent transactions. The assistant could provide an analysis without needing permission to transfer funds or make purchases.
Read-only access does not eliminate every privacy or security concern. Financial information can still be sensitive, and users should understand what data is shared, how permissions work and how connected services handle that information. However, restricting an agent to viewing information can reduce the potential damage caused by an incorrect instruction or unauthorised purchase.
Anthropic Is Also Moving Towards More Restricted Financial Access
Google is not alone in exploring safer ways to connect AI assistants with personal finances. Anthropic reportedly introduced a similar read-only approach earlier in November, reflecting growing interest in separating financial analysis from the ability to execute transactions.
The distinction is important as AI companies compete to develop more capable agents. A system that can examine personal data, understand spending patterns and provide recommendations may be useful without needing complete control over the accounts involved.
For example, someone trying to manage a monthly budget could ask an AI assistant to:
- Summarise spending across different categories.
- Identify recurring subscriptions and regular payments.
- Compare spending between different months.
- Highlight unusually large transactions for review.
- Identify potential opportunities to reduce expenses.
- Prepare a budget based on available transaction information.
These activities generally require access to financial information rather than the ability to move money. A read-only connection can therefore offer useful assistance while limiting the agent’s ability to take direct financial action.
Read-Only Access vs Full Banking Access: What’s the Difference?
The difference between these two approaches comes down to permissions. A read-only tool can examine information, while a more powerful financial agent may be able to interact with services and initiate actions.
| Feature | Read-only AI access | Broader AI account access |
|---|---|---|
| Review transactions | Yes, where supported | Yes |
| Analyse spending patterns | Yes | Yes |
| Identify recurring payments | Yes, where data is available | Yes |
| Prepare financial recommendations | Yes | Yes |
| Initiate purchases | Not through read-only permissions | Potentially, depending on granted permissions |
| Make account changes | Not through read-only permissions | Potentially, depending on the connection |
| Risk from incorrect transactions | Lower direct transaction risk | Higher if transaction permissions are enabled |
Read-only access is not necessarily a complete solution, but it follows an important security principle: give software only the permissions it needs to perform its job.
For many personal-finance tasks, the ability to inspect and organise information is enough. Users who want an AI assistant to recommend ways to save money may not need to let it make purchases or move funds.
What Happens If an AI Agent Gets Hacked?
The possibility of an AI agent being compromised is one of the more serious concerns behind the banking changes. An agent connected to financial accounts can become an attractive target because it may hold permissions that allow it to interact with sensitive services.
Consider a situation in which a customer connects an AI assistant to a bank account and later loses control of the assistant’s account. If an attacker gains access to the agent and can use its existing permissions, they may attempt to perform actions through the connection.
That creates a complicated chain of responsibility. The bank may argue that the customer authorised the third-party service. The AI provider may point to the permissions granted by the user or the security of the connected account. The customer, meanwhile, may argue that the resulting activity was not personally approved.
This is precisely why the banks are making their positions clearer in their terms of service. They want customers to understand that connecting an external AI tool is not necessarily the same as using the bank’s own application.
Customers should also remember that a read-only connection can still expose sensitive financial information if the AI account or connected service is compromised. Restricting transaction permissions reduces one category of risk, but it does not remove the need to protect personal data.
How to Use AI Assistants More Safely With Your Bank Account
AI-powered financial tools can be useful, but customers should be careful about how much access they provide. A few basic precautions can reduce the chances of an automated mistake turning into a financial problem.
- Avoid sharing banking passwords: Do not give an AI assistant your banking username, password or one-time authentication codes unless the bank explicitly supports a secure connection method.
- Prefer official integrations: Use connections provided through recognised financial platforms rather than entering credentials into unfamiliar third-party services.
- Choose read-only access where possible: If your goal is to analyse spending or track payments, avoid granting permission to initiate transactions.
- Review permissions regularly: Check which applications and services can access your financial information and revoke connections you no longer use.
- Require confirmation for payments: Keep human approval in the process before an AI agent makes purchases, transfers or other financial commitments.
- Enable account alerts: Transaction notifications can help you identify unexpected activity quickly.
- Protect your AI account: Use a strong, unique password and multi-factor authentication wherever supported.
- Read updated bank agreements: Understand how your bank treats activity performed through third-party applications and AI agents.
If you notice a transaction you did not approve, contact your bank immediately through its official support channels. Do not assume that an AI-related dispute cannot be challenged simply because the agent had previously been granted access.
Will Banks Eventually Block AI Agents Completely?
The current changes do not necessarily mean that banks intend to prevent every form of AI interaction with financial accounts. Instead, they suggest that financial institutions want clearer limits around third-party access and responsibility.
AI assistants are becoming more capable of performing tasks that previously required a person to navigate websites and applications manually. Banks may eventually support more secure ways for customers to use those capabilities without sharing passwords or granting unrestricted account access.
One possible direction is permission-based integration, where customers can allow an AI tool to view selected information or perform narrowly defined actions. Such systems could provide more control than simply handing an agent access to an entire account.
Another important development will be transaction-level confirmation. Rather than allowing an AI assistant to make payments whenever it decides an action is necessary, financial services could require explicit approval for each transaction or for actions above a certain limit.
However, the technical and legal questions are not yet fully settled. Banks, AI companies and regulators will need to address authentication, fraud investigations, data protection and liability as automated financial services become more common.
The Bigger Question: Who Is Responsible When AI Controls Your Money?
The banking changes highlight a growing tension in the AI industry. Technology companies are building assistants that can do more than answer questions, while financial institutions are trying to prevent those same systems from creating new security and liability problems.
For customers, the attraction is understandable. An AI agent that can examine spending, organise bills and handle routine purchases could save time. But giving software access to a bank account is fundamentally different from asking it to summarise an article or draft an email.
When an AI assistant makes a mistake involving money, the consequences can be immediate. A wrong recommendation may be inconvenient, but an incorrect payment or unauthorised account action can create a financial dispute that takes time to resolve.
Read-only integrations from companies such as Google and Anthropic offer one possible compromise. They allow AI systems to provide financial insights without automatically giving them the power to spend. For many users, that may be the more sensible starting point.
The message from major US banks is increasingly clear: customers should understand the risks before connecting third-party AI agents to their accounts. As automated assistants become more powerful, convenience alone is not a sufficient reason to hand over control of personal finances.
For breaking news and live news updates, like us on Facebook or follow us on Twitter and Instagram. Read more on Latest Technology on thefoxdaily.com.

COMMENTS 0