On October 27 alone, several moves to rein in AI agents followed one after another. Canada's RBC joined a trend among banks of blocking access by AI agents. Oracle built a method for assigning each agent an identity into the "AI Workspace" it is developing with OpenAI. SK Telecom published its first governance report addressing the question of accountability for agents.
An AI agent is an AI that, once a person hands it a goal, carries out multiple steps on its own, from searching and clicking to making payments. It is convenient. It also comes at a cost. A piece of industry content released on October 26 named autonomous agents that act beyond the scope of their instructions as one of the headaches facing the AI industry, alongside chip shortages and tighter scrutiny from regulators.
A program that does more than it was told to can lead to operational accidents when used in actual work.
Banks are taking the firmest stance. Which channels were closed, and which banks moved first, has not been disclosed. If a bank cannot tell whether a connection to an account comes from a human hand or a program, locking the door is the easiest response.
Oracle chose to keep the door open while handing out access badges. Giving agents an identity allows a company's systems to distinguish each agent the way they would a human employee, and to set separately what information it can see and what tasks it can perform. Oracle also said it has built security features into its database. At the announcement, AI was even likened to "a teenager who is hard to control."
The SK Telecom report deals with accountability instead of technology. Governance refers to a framework that sets out who manages AI under what rules, and who answers when an accident occurs.
Responses vary from company to company, and no common standard shared across the industry is in sight. There were also calls to make agent access control the top priority and to shorten the time it takes from detecting a risk to blocking it.
Kuntech forecast that a market for controlling agents at the operational stage will take shape in 2027. Assistant AI that opens banking apps on a user's behalf will be difficult to use at banks like RBC for the time being. Once company agents are given identities like employee numbers, it will become possible to trace, one by one, which agent did what.
