Major banks are restricting or blocking third-party AI agents from accessing customer accounts, according to Global Finance. The issue is not just what an agent can do, but how a bank can tell whether a customer authorised it to do so.
Watch Desk analysis
What happened
A Global Finance article published on 28 October says JPMorgan Chase, Bank of America, Citigroup, Wells Fargo and Morgan Stanley have restricted or blocked third-party agents’ access, citing difficulties distinguishing authorised activity from actions by rogue agents. The article describes reported incidents of agents gaining unauthorised access to customer bank accounts, but does not detail those incidents.
Andrew Shikiar, chief executive of the FIDO Alliance, argues that verifiable digital credentials could let an agent identify the person who authorised it and the limits of that authority. The article also sets out a practical control for businesses: reassess controls when they change an AI model, connect a new tool, alter a prompt or broaden an agent’s mandate.
Why it matters
For banks, an agent that can initiate payments is also an identity and authorisation problem. If a system cannot reliably establish who sent an instruction, what they permitted and whether the agent stayed within those limits, restricting access is a straightforward response. For organisations deploying agents, a model update or new tool connection may need to trigger a fresh check of the full action sequence, not just the individual steps.
Our read
The banks’ restrictions are a sensible stopgap, not a workable end state for useful banking agents. Digital credentials and better audit controls could help, but the important test is whether permissions remain clear and enforceable as an agent moves between systems. A digital ID is not a tiny responsible adult inside the machine.
What to watch
- Whether banks keep restricting agent access or begin allowing it under defined controls.
- Whether digital credentials can show an agent’s identity and specific authorised actions.
- How organisations test an agent’s complete execution path after changes to its model, tools or mandate.
Discussion spark: Should banks keep third-party AI agents out until their identity and permissions can be verified, or would that shut down useful automation for too long?
Sources and evidence
- Source update (28 October 2026, 17:31 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.