These are generally situations when an agent or AI assistant accesses a human's account with the user's permission and login credentials. Even if permitted by the user, this goes against the guidelines of most banks requiring that humans log in to their own accounts.
The mind boggles...
Whoever gives AI their banking credentials ain't right in the head.
These fucking companies are not only borrowing and spending such obscene amounts of money they're going to crash the economy, they're also forcing everyone else to spend huge amounts of their own time and money to try and deal with their activity.
From banks, online vendors and other businesses, to websites constantly being scraped and effectively DDOS'ed, to software project maintainers being inundated with pull requests, these things are making all our lives hell.
Sounds like an ideal application area for prompt injection.
I bet there are folks out there setting up agents like "hey robot, log in and pay all my bills" and just turning it loose to be lazy
"some agents attempting to take actions that appear to be independent of user instructions"
appear to be?
no source?
also, no victims?
Language models do not act on their own,
and humans are after your money, programs aren't.
Either this didn't happen or someone is doing this, using agents as tools.
When we were scared of god, it didn't benefit god, it benefited the people who created it.
LMBO what dumbo would give A"I" access to their money?
It would be really funny to see if the chatbots are only accessing accounts that have bad passwords like "password" because they essentially guessed it by accident.
all 13 comments