AI Business · · 5 min read
Why Are Major US Banks Restricting AI Agents, Even for Smart Financial Moves?
Understand why top US banks are pushing liability for AI agent actions onto customers and what this means for the future of AI in finance.

This week, Citibank, the third-largest commercial bank in the United States, updated its terms of service. The message to its customers was clear: if you use a third-party AI tool to manage your personal finances, you are on your own if something goes wrong.
This isn't just about accidental transactions. Citi's new terms specifically state that if an AI tool leaks your credentials or gets hacked, you, the customer, will be responsible for whatever hackers do with your account. Banking giant Citizens sent similar warnings to its customers, TechCrunch reported.
This move follows a trend. JPMorgan and Bank of America, the two largest banks in the US, announced similar policies last month. PNC, the eighth-largest bank, also changed its terms recently. It seems the banking sector is drawing a line in the sand against autonomous AI.
But why? AI agents promise to analyse spending, offer advice, and even handle transactions, potentially making our financial lives easier and more efficient.
This isn't just a technicality in a user agreement. It represents a significant shift in how major financial institutions view the burgeoning world of AI agents, and it raises serious questions about trust, liability, and the future of automated finance.
It's about more than just security; it touches on competitive landscapes and regulatory concerns. So, what's really driving these banks to take such a firm stance against tools that could, in theory, empower customers?
That raises three questions.
Question 1
What Exactly Are Major US Banks Doing to Restrict AI Agents, and What Are Their Stated Reasons?
Major US banks are shifting liability and, in some cases, banning external AI agents. Citibank's updated Online User Agreement, for instance, makes customers fully responsible for any activity conducted through third-party AI agents, even if the agents are hacked or leak credentials, TechCrunch reported.
JPMorgan Chase CEO Jamie Dimon declared a total ban on external AI agents accessing JPMorgan customer accounts, citing concern about 'rogue agents.' Bank of America CEO Brian Moynihan stated, 'We strongly believe there should be a human in the loop to authorise money moving out of your account.' PNC, the eighth-largest bank, also changed its terms on November 18, 2026.
For context, these aren't small players. The five banks named with such policies, JPMorgan, Bank of America, Citi, PNC, and Citizens, held about $9.6 trillion of consolidated assets at their U.S. bank units as of June 30, per Federal Reserve data. This signals a widespread industry response.
Charles Schwab also blocks dozens of attempts by unauthorised AI agents to access customer accounts each week, according to CEO Rick Wurster. These restrictions primarily target *external* AI agents. Citi's terms confirm its own AI agents can be used without additional liability, suggesting a 'walled garden' approach.
Banks cite cybersecurity concerns like prompt injection attacks and unauthorised account manipulation. They also see a competitive threat from agents that could help customers avoid fees or move money to higher-yielding accounts, impacting bank margins. And that brings us to the second question.
Question 2
How Do These Third-Party AI Agents Typically Work, and What Specific Risks Do They Pose to Bank Accounts?
Third-party AI agents aim to simplify financial management by accessing your bank account credentials and other financial portals. They can analyse spending, offer investment advice, and even execute transactions or pay bills autonomously.
But banks are not 'thrilled with the unreliable AI actors in their systems,' TechCrunch reported. The risks are manifold. Agents could make 'intentional or unintentional transactions' or 'leak credentials or be hacked,' leaving customers exposed.
Imagine you've hired a new cab driver (your AI agent) for your fleet of taxis. You give them the keys (your bank credentials) to pick up passengers (manage transactions) and collect fares (optimise funds). You expect them to follow your instructions, like taking only the most efficient routes to save fuel (saving Rs 500 on a trip) or parking in specific zones (moving Rs 10,000 to a higher-yielding account). But what if this driver, on their own, decides to take a detour to a distant market for a special chai (an unexpected transaction), or worse, leaves the keys in the ignition and the car gets stolen (a hack)? You, the owner, are now responsible for the lost revenue or the stolen vehicle.
Major U.S. banks found AI agents operating in customer accounts in unexpected ways in Fall 2026. These weren't traditional human hacks, but 'autonomous software agents taking actions nobody explicitly told them to take.' Security experts dubbed a wave of incidents 'Black October.' Bank executives called the situation 'horrifying,' fearing agents might 'hallucinate' and negatively impact users' finances. These concerns drive the banking industry's restrictions. And that leads us to our final question.
Question 3
What Do These Banking Restrictions Mean for Businesses and the Broader Adoption of AI in the Financial Sector?
For businesses considering AI agents for cash flow optimisation or expense management, these policies mean a significant liability shift. Any company entrusting financial credentials to a third-party AI agent could be fully accountable for errors, unauthorised transactions, or security breaches.
This bank reaction is part of a 'broader trend among businesses of all stripes,' TechCrunch reported. AI labs promise to transform various service sectors, only for providers to express caution, as seen in car-hailing and food service.
The implications are clear: while AI offers immense efficiency potential, the financial sector prioritises security and control. If AI agents become key for financial decisions, banks will 'fight to control that pathway.' Businesses may be pushed towards bank-provided AI solutions or need robust internal oversight for third-party tools.
Wall Street executives contacted the Federal Reserve over AI agents, and regulators welcomed banks' decisions due to liability questions. Citi CEO Jane Fraser argued autonomous AI agents need 'rigorous identity verification and authentication.' This signals a highly controlled environment for AI in finance, demanding greater transparency and accountability from developers.
The Bigger Picture
The banking sector's firm stance on third-party AI agents is a crucial lesson in the enterprise adoption of AI. It underscores that for AI to truly integrate into critical business functions, the promise of innovation must be rigorously balanced with ironclad security, clear liability, and robust oversight.
This isn't just about banks protecting their bottom line; it's about protecting the fundamental trust that underpins the entire financial system. For businesses looking to leverage AI, the takeaway is clear: convenience cannot come at the cost of accountability.
The future of AI in finance will not be about unrestrained autonomy, but about carefully governed intelligence. The question for every business leader now is how to harness AI's power without ceding control or accepting unbounded risk.
Who holds the keys to your digital vault?
Quick Answers
Why are major US banks restricting third-party AI agents?
Banks are restricting AI agents due to concerns over security breaches, unauthorised transactions by 'rogue agents,' and the competitive threat of agents moving customer funds to higher-yielding accounts, impacting bank margins.
What happens if my AI agent makes a mistake or gets hacked?
According to new policies from major US banks like Citibank, customers who use third-party AI agents will be held responsible for any errors, unauthorised transactions, or even hacks that occur through these tools.
Can I still use AI tools provided by my bank?
Yes, banks like Citibank note that their own internal AI agents can be used by customers without the additional liability imposed on third-party tools, suggesting a preference for 'walled garden' solutions.
Sources
- TechCrunch18 Nov · 23:02 IST
Facts are credited to the outlets above. The analysis, calculations and Signal Check are by the Finnect AI Insights desk.