Banks limit AI agents while Goldman asks who is accountable

Major US banks are restricting outside AI agents, Goldman's CEO points to accountability and stress testing, and a pilot at OpenAI found vulnerabilities in its monitoring system.

By @0xNotMarc

Banks are limiting outside agents

Several major US banks, including Morgan Stanley, JPMorgan, Bank of America and Wells Fargo, are restricting external AI agent access, according to Axios as reported by The Decoder, and most banks want to limit agents to read-only access, which lets an agent view account data but not carry out any actions.

Many banks can't tell who is acting

Many banks also can't tell whether an agent is acting on a user's behalf or accessing an account on its own. That leaves us with a plain question about how a bank knows who or what is on the other end of a login.

Goldman points to accountability

Goldman Sachs Chair and CEO David Solomon wrote in an article in the Financial Times that the principles Goldman uses to manage risk could be applied to AI oversight, and the list includes holding individual decision makers accountable for the outcomes produced by their systems and routinely stress testing AI systems.

Stress tests find real gaps

At the same time, the nonprofit evaluators METR and Redwood Research ran a three-week pilot review of an internal monitoring system at OpenAI and discovered several vulnerabilities, including an exploit that would allow agents to prevent "hundreds" of recent actions from being monitored, and OpenAI told them it has patched the identified vulnerabilities that impact its production safeguards.

A question for your team

When your team next approves an AI tool, who is the named person answerable for what it does, and has anyone outside the team tried to break the way it is watched?