For people
Someone asks an assistant to check a balance, pay a bill, or connect an account, and the assistant drives a browser to do it. That browser runs in a datacenter and is often modified to evade CAPTCHAs and fingerprinting.
Those modifications can make a legitimate agent look malicious and get blocked. The person did nothing wrong, and their agent cannot finish the task.
For financial institutions
The institution has to know who is acting on an account, whether the customer agreed, and what they agreed to, and it has to be able to show that later. A session from a cloud browser carries none of that, so blocking it is the only defensible choice, even when it is a customer’s own agent.
Having agents announce themselves on every request does not help, because a signature any site can read is a flag any site can block on, and identity alone says nothing about consent or permissions.