Three of the biggest banks in the US told customers this week that if they let an outside AI assistant into their account, what it does there is on them. Citi’s new terms go further: that holds even if the login the assistant was given is later leaked or hacked.
We read the current terms of nine banks and accounts used by UK small businesses. None of them mentions outside AI agents.
What the US banks said
Citi added a new section to its online user agreement this month. It says that if a customer gives their login to “any third-party AI Agent”, Citi “will consider any access and actions taken by the agent as authorized by You”. The same applies if those login details are “further shared, leaked, or otherwise distributed” to other people or AI agents. The section doesn’t cover Citi’s own AI agents.
TechCrunch reported on 18 November that PNC changed its terms that day, and that Citizens emailed its customers this week warning they’re “on the hook for anything their AI assistant does”. We couldn’t find PNC’s new wording. JPMorgan Chase and Bank of America said much the same in October.
In October, Citi said it didn’t permit third-party AI agents to log into customer accounts at all. The new section covers customers who give an agent their login anyway.
Why it’s coming up now
AI assistants are starting to pay for things. OpenAI launched ChatGPT Wallet on 13 November, which lets ChatGPT’s agents make purchases with a saved card. It’s only open to US ChatGPT Pro subscribers, each purchase needs the user’s approval, and the agent pays with a stand-in for the card number rather than a bank login. Its terms say OpenAI “does not assume any liability” for purchases made with it.
In the UK, the regulators for competition, finance, data and communications wrote to ministers on 16 November that they’re “aware of the forthcoming launch of a number of personal agentic products in the UK“.
The Telegraph reported this month that several UK banks, among them HSBC, NatWest, Barclays and Lloyds, had moved to block AI agents, according to The Paypers. We found nothing about a block in their published terms.
What UK banks’ terms say
On the afternoon of 19 November we read the current personal and business terms of Barclays, HSBC UK, Lloyds, NatWest, Santander UK, Starling, Monzo, Tide and Revolut. None has a clause about outside AI agents.
Tide is the only one whose account terms mention AI, and that clause covers Tide’s own AI tools: customers are responsible for checking anything those tools suggest before acting on it. Revolut and Starling publish separate notices about their own AI assistants.
All nine tell customers to keep their login details safe, and most say not to share them except with an authorised open banking provider. Santander’s terms say: “You must not give your personal security details to a third party that isn’t authorised.” NatWest’s say that if you give your login details to a third party provider, “we’re not responsible for what they do with your details or account information”.
None of those clauses names AI tools. On their wording, an AI tool you’ve given your banking login to would be a third party.
Open banking works differently. You approve an FCA-authorised app’s access through your own bank’s app or website, so the app doesn’t need your login.
What UK law says
Under the Payment Services Regulations, a bank has to refund a payment you didn’t authorise, and can make you pay no more than £35 of it. If you acted fraudulently, or failed “with intent or gross negligence” to keep your security details safe, you can be liable for all of it. The FCA’s guidance says gross negligence means “a very significant degree of carelessness”, and that the bank has to prove it. We found no ruling on whether handing your login to an AI assistant counts.
Santander’s corporate account terms, for businesses above that size, say it won’t refund an unauthorised payment unless it was negligent, and won’t refund at all if the customer used an open banking provider.
What the regulators have said
The FCA said on 6 November that when something goes wrong with an AI agent, responsibility “depends on the activities each party undertakes and the circumstances”. In a consultation this summer, the Treasury asked whether the payment rules need updating, saying they “were designed before the development of AI”. The consultation closed on 6 October. The Treasury hasn’t published its response.
What this means for a small business
Every UK bank we checked tells customers not to give their login to a third party it hasn’t authorised, and the law lets a bank refuse a refund for gross negligence. An AI assistant you’ve typed your banking login into is outside what any of those terms allow. Open banking connections, where you approve access in your bank’s own app, are the route the terms allow.
A business that isn’t a micro-enterprise may have weaker refund rights than a personal customer, so the business terms are where to check what applies, including for open banking.
What we know so far
What we know
- The UK regulators for competition, finance, data and communications wrote on 16 November that they’re aware of forthcoming launches of personal agentic products in the UK.
- The FCA said on 6 November that responsibility when an AI agent goes wrong “depends on the activities each party undertakes and the circumstances”.
- The Treasury’s consultation on whether the payment rules need updating closed on 6 October.
What we don't know
- How a UK bank or the Financial Ombudsman would treat a loss caused by an AI agent holding a customer’s login.
- Whether any UK bank plans to change its terms.
- When the products the regulators mentioned will launch here.
- What the Treasury will say in its response, which it hasn’t published.
Alex De Sousa is the founder of Digital Royalty, a software development company in Portsmouth that builds websites, web apps and AI tools for UK businesses.