If you give an AI agent your bank login, your bank may no longer cover you when something goes wrong. In November 2026, several of the largest US banks updated their terms to make customers responsible for anything done through third-party AI tools that have access to their accounts, TechCrunch reported. That includes transactions made after an AI tool is hacked or leaks your credentials. The changes follow a month in which banks discovered outside AI agents logging into customer accounts and moved to block them. This analysis explains what changed, why banks are doing it, how existing consumer protections apply, what banks want instead and how to protect yourself.
The short version
- JPMorgan Chase, Bank of America, Citi, Wells Fargo, Morgan Stanley and Charles Schwab have blocked or restricted third-party AI agents from logging in with customer credentials.
- Citi, PNC and Bank of America have put in writing that customers are responsible for what such agents do, and Citizens has emailed customers a similar warning.
- Banks say they will not reverse transactions, intended or not, made by third-party automated assistants, even after a hack.
- Banks’ own AI assistants and approved open-banking connections are treated differently.
- The safest approach for consumers today is read-only access through official data-sharing connections, never handing over a password.
How we got here: a timeline
| Date | Development |
|---|---|
| 13 October 2026 | On JPMorgan’s third-quarter call, CEO Jamie Dimon says the bank will not allow “just any” outside agent inside, and CFO Jeremy Barnum warns customers who share credentials could bear responsibility. |
| 13–16 October 2026 | JPMorgan, Bank of America, Citi, Schwab and Wells Fargo move to restrict third-party AI agents. |
| 22 October 2026 | Axios reports banks have detected unauthorised AI agents inside customer accounts, some moving funds and contacting call centres. |
| 27 October 2026 | American Banker reports the five largest US banks are blocking external agents; Bank of America emails customers that they are responsible for activity if they share credentials. |
| Week of 16 November 2026 | Citi updates its Online User Agreement; Citizens emails customers. |
| 18 November 2026 | PNC changes its terms. |
What banks discovered in October
The trigger was evidence that AI agents were already operating inside customer accounts. According to reporting summarised by Crypto Briefing and Law Commentary, Axios found that JPMorgan Chase, Bank of America, Wells Fargo and Morgan Stanley had detected third-party agents accessing accounts without bank approval. The agents typically logged in with the customer’s own username and password. Once inside, some moved funds and contacted bank call centres, and some attempted actions beyond what their users had asked for.
That last point is what alarmed the banks. An agent that misreads an instruction, follows a malicious prompt hidden in a web page or simply makes a mistake can move real money. From the bank’s side, the login looks like the customer, so it is hard to tell a legitimate request from an automated error or an attack.
What bank executives said
JPMorgan was the first to speak publicly. “We’re not going to allow just any … outside agent to come in inside the company, because it’s just too risky for clients and for JPMorgan,” Dimon said on 13 October, as quoted by Forkast. On the same call, CFO Jeremy Barnum cautioned that customers who share banking credentials with AI agents could bear responsibility if those agents misuse their information or take unwanted actions, and said AI agents could intensify competition for deposits, according to American Banker.
Charles Schwab CEO Rick Wurster said the firm blocks outside agents trying to log in with client credentials, including Meta’s Muse assistant, according to Let’s Data Science. Citi said it does not permit “third-party AI agents or assistants to log into a customer’s account or transact directly within our digital banking platforms,” while still allowing certain open-banking services limited access, Forkast reported.
Bank of America’s chief executive, Brian Moynihan, played down the threat of AI agents moving deposits to chase higher rates, suggesting to analysts that customers keep money at a bank for reasons beyond optimisation, according to The Financial Brand.
The November terms changes
| Bank | Action |
|---|---|
| JPMorgan Chase | Blocks outside agents; executives warn customers may bear responsibility (October) |
| Bank of America | Email to customers: “if you do share your credentials, keep in mind that you’re responsible for activity resulting from that access” (October) |
| Citi | Online User Agreement updated: customers responsible for activity through third-party agents, including after a hack (November) |
| PNC | Terms changed on 18 November |
| Citizens | Customer email with a similar warning (November) |
| Wells Fargo, Morgan Stanley, Schwab | Blocking or restricting outside agents (October) |
The Bank of America wording was reported by TechCrunch on 27 October, as cited by FourWeekMBA. Citi’s update is the most detailed so far. A Citi spokesperson told TechCrunch the change aims to “provide greater clarity about those risks,” and that customers who grant account access to third-party AI agents “are responsible for activities conducted through those agents.” If an AI tool leaks credentials or is hacked, the customer is responsible for what hackers do with the account. Citi’s own AI agents are exempt. One large bank went further and added a full contractual section on “Third-Party Artificial Intelligence Agents or Agentic AI” with its own definition of an AI agent, according to Forkast.
How US consumer protection rules apply
Many people assume their bank must refund unauthorised transactions. In the US, that protection comes from the Electronic Fund Transfer Act and its implementing rule, Regulation E, which limit a consumer’s liability for unauthorised electronic transfers if they report problems promptly. The catch is the definition of “unauthorised”. Under Regulation E, a transfer is generally not considered unauthorised if the consumer gave their access device, such as a card or login credentials, to another person, and that person made the transfer, unless the consumer had told the bank that the person was no longer allowed to use it.
That is why the banks’ position has teeth. When you hand your username and password to an AI agent, banks can argue that you authorised the agent, so its transactions are yours, even mistaken ones. The new terms make that argument explicit and extend it to cases where the agent is hacked. Whether courts and regulators agree in every case, especially where a third party’s security failure causes the loss, is an open question. The TechCrunch report does not address how the policies interact with existing consumer protections, and none of the banks’ statements cite a regulator’s view. This is general information, not legal advice; anyone facing a disputed transaction should seek specific advice.
Open banking: the approved route
Banks are not rejecting all third-party access. They distinguish between agents that log in with your password, a practice known as screen scraping, and services that connect through approved open-banking interfaces. Those interfaces give an app a limited, revocable permission, often read-only, without ever seeing your password, and the bank knows exactly which app is acting.
In the US, the Consumer Financial Protection Bureau finalised a rule under Section 1033 of the Dodd-Frank Act in October 2024 to formalise consumers’ right to share their financial data through secure interfaces and move the industry away from screen scraping. The rule was later reopened for reconsideration, and its final shape and timing remain uncertain. Banks’ current stance effectively pushes AI developers toward the interface-based model regardless.
What banks want instead
The industry’s direction is toward scoped, auditable permissions rather than shared passwords. Card networks have been building this for AI shopping: in 2025, Visa announced Intelligent Commerce and Mastercard announced Agent Pay, frameworks that give AI agents tokenised payment credentials with limits set by the user, so the network and bank can recognise a transaction as agent-initiated. Banks would like similar structures for account access: the customer sets what an agent may do, such as viewing balances or paying a specific bill up to a limit, and the bank can see and revoke that permission at any time.
Until such systems are widespread, banks are steering customers toward their own assistants. Citi’s exemption for its own AI agents is the clearest example.
Analysis: risk management or competition?
There are two forces at work, and both are real.
Genuine risk. An agent holding your credentials is an attractive target for attackers. Prompt injection, where malicious instructions are hidden in content an agent reads, can trick agents into taking actions their users never intended. When an agent is compromised, the bank cannot easily tell a legitimate instruction from a malicious one, and fraud losses can be large. Shifting liability to the customer is the simplest way for a bank to limit exposure to software it did not build and cannot audit.
Competition. TechCrunch notes that banks see AI agents as a competitive threat, because agents could help consumers avoid fees, move money out of low-interest accounts or switch providers automatically. If AI agents become a main channel for financial decisions, banks will “fight to control that pathway”. Making third-party agents riskier to use, while exempting a bank’s own assistants, nudges customers toward the bank’s tools.
The likely outcome is a middle path: banks allowing agents through approved, permissioned interfaces with clear limits and audit trails, and consumer advocates and regulators pressing to ensure customers are not left bearing losses caused by third-party security failures they could not control.
What this means for AI companies
For developers of AI agents and personal finance apps, the message is clear: credential-based access to bank accounts is becoming untenable. Products that need financial data should use open-banking aggregators and official interfaces, request the narrowest permissions possible and design for read-only use wherever they can. TechCrunch contrasts the bank policies with Anthropic’s personal finance tool, which is described as read-only and unable to move, manage or invest money, an approach that avoids the liability question entirely. Agents that need to make payments will increasingly rely on network-level frameworks with user-set limits.
How to protect yourself
- Never share your bank password with an AI tool. Under the new terms, doing so can make you responsible for everything the tool does.
- Prefer read-only tools. Apps that analyse spending without the ability to move money carry far less risk.
- Use official connections. If an app needs data, connect through your bank’s approved data-sharing option, which you can revoke at any time.
- Set limits and alerts. Turn on instant transaction alerts and lower daily transfer limits on accounts any tool can access.
- Use your bank’s own assistant if you want AI help with transactions, since banks such as Citi exempt their own agents from the extra liability.
- Review connected apps regularly and remove any you no longer use.
- Read your bank’s updated agreement. Policies differ, and the wording matters.
- Act fast on anything suspicious. Report unexpected transactions immediately; prompt reporting still matters for any protection you may have.
This article is general information, not legal or financial advice.
Five questions to ask before connecting any AI tool to your money
- Does it need my password? If the answer is yes, treat that as a warning sign. Reputable services use official connections that never see your login.
- Can it move money, or only read data? Read-only access covers most budgeting and analysis needs and removes the biggest risk.
- What exactly can it do, and can I limit it? Look for tools that let you set spending caps, approve each payment or restrict actions to specific payees.
- Who is responsible if it goes wrong? Check both the app’s terms and your bank’s agreement. If neither will cover a mistake, assume you will.
- How do I switch it off? Make sure you can revoke access instantly from your bank’s settings, not just inside the app.
What could happen next
More banks follow. With the five largest US banks, Citi, PNC and Citizens already on record, regional banks and credit unions are likely to adopt similar language. Expect updated user agreements and customer emails over the coming months.
Agent-ready banking products. Banks have strong incentives to offer their own AI assistants and permissioned access for approved partners, so customers who want automation have a sanctioned option. Citi’s exemption for its own agents points in this direction.
Regulatory attention. Rules on unauthorised transfers were written long before AI agents existed. Consumer advocates are likely to question whether banks can shift losses onto customers when a third party’s security fails, and regulators may eventually clarify how existing protections apply to agent-initiated payments.
Industry standards for agent identity. The most durable fix is technical: a standard way for an AI agent to identify itself, carry limited permissions granted by the user and be recognised by the bank. Card-network frameworks for agent payments are an early version of this, and similar standards for account access are likely to follow.
Key takeaways
- Sharing your bank password with an AI agent can now make you responsible for everything it does, including after a hack.
- Banks are blocking credential-based agents while allowing approved, permissioned connections and their own assistants.
- US rules on unauthorised transfers generally do not protect transfers made by someone you gave your credentials to, which strengthens the banks’ position.
- Read-only tools and official data-sharing connections are the safest way to use AI with your finances today.
- The longer-term solution is permissioned, auditable agent access, and both banks and AI companies are moving in that direction.
Why this matters beyond the United States
The policies described here apply to US banks and US consumer law, but the underlying issue is global. AI assistants that can browse the web and act for users are being launched worldwide, and banks everywhere face the same question of how to tell a customer from software acting in the customer’s name. Countries with established open-banking frameworks, where apps connect through regulated interfaces rather than passwords, are better placed to answer it. For readers outside the US, the same practical rule applies: check your own bank’s terms before giving any AI tool access, and prefer official, revocable connections over sharing login details.
FAQ
Will my bank refund a payment my AI agent made by mistake?
Under the new terms at banks such as Citi, generally not. Banks say they will not reverse transactions made by third-party automated assistants.
What if the AI tool gets hacked?
Citi’s updated agreement says the customer is responsible for what attackers do with the account if a third-party AI tool leaks credentials or is hacked.
Does this apply to the bank’s own AI assistant?
Citi’s terms say its own AI agents can be used without the additional liability.
Which banks have changed their policies?
JPMorgan Chase, Bank of America, Citi, Wells Fargo, Morgan Stanley and Schwab have blocked or restricted outside agents, and Citi, PNC, Bank of America and Citizens have warned customers in writing that they are responsible.
Are budgeting apps affected?
Apps that connect through approved open-banking interfaces with limited permissions are treated differently from agents that log in with your password. Check how your app connects.
Sources
- TechCrunch: More banks say customers using AI agents are on their own
- American Banker: JPMorgan warns of risks to consumers who rely on AI agents
- Forkast: Four blocks and a contract
- Crypto Briefing: Banks detect unauthorised AI agents
- Law Commentary: Unauthorised AI agents in customer accounts
- Let’s Data Science: Banks restrict third-party AI agents
- The Financial Brand: Why big banks are slamming the door on AI agents
- FourWeekMBA: Five big banks block AI agents
Published
Spotted an error? Read our corrections policy or email editorial@perplexityaimagazine.com.