브레스저널 The Breath Journal

This article was translated automatically from the Korean original. Read the original in Korean

From 89 People to the Entire Financial Sector, the Cost of Attack Has Collapsed

곽동현·Published 2026-10-04 12:00 KST
Indiscriminate Scanning in the Hands of AI Agents, Regulators Call Emergency Meeting
Attacks that once aimed at a single target have shifted to sweeping the entire grid
Attacks that once aimed at a single target have shifted to sweeping the entire grid / ⓒ Breath Journal

The personal information of 89 people was taken from Hana Bank in a hack believed to have used an AI agent. The figure, released on October 2, 2026, is small in itself. But in the same week a security hole at Shinhan Bank was exposed to a similar form of attack, and financial regulators held an emergency meeting on October 4 to activate an emergency response system. Regulators do not view this as an incident at one or two companies.

The reason a number as small as 89 moved regulators is that the arithmetic of attack has changed.

Hacking used to be the work of people. Attackers picked their targets after weighing which company to go after and whether the time invested matched the odds of success. Large financial firms were therefore worth investing in defenses, and small financial firms were somewhat safe simply by being less visible.

An AI agent is software that takes a goal in place of a person and breaks it into steps and executes them on its own. Once an attacker has one, regulators say, it becomes possible to sweep externally exposed systems en masse without discriminating among them.

When selection disappears, the assumption that "we are safe because we are small" disappears with it. That is the background to regulators placing more weight on an indiscriminate offensive across the whole financial sector than on a localized breach aimed at a particular company. Indications have also emerged that the targets include not only major commercial banks but savings banks, capital firms and mutual finance institutions. How many firms in each sector were affected will only take shape once the regulators' investigation is complete.

Regulators' view that existing security capabilities cannot block new types of attacks using AI falls in the same line. Intrusion detection systems are generally designed to catch "one suspicious access." When the opponent is an attack sweeping many places at once, people cannot keep up even when alarms go off.

Industry had similar worries. In a PwC survey of 3,934 corporate and technology executives in 71 countries from May to July 2026, half of the executives responsible for security and technology named attacks aimed at AI systems as the threat they are least prepared for. At the top of the threat list, alongside botnets, were adversarial attacks that feed in inputs designed to deceive AI models, and data poisoning that mixes corrupted values into training data to make a model reach the wrong conclusion.

There is movement on the supply side as well. On October 3 Apple changed the Mac to require stricter conditions for access to personal information, warning of the risks AI agents could bring, and Nvidia released a safeguard the same day that cuts off agents that go beyond control. Neither measure was described as directly linked to the situation in Korea's financial sector, but both the platform and the chip side have begun reworking the scope of permissions to be granted to agents.

Much remains unresolved. When the attack actually began, whether it is still continuing, and what technical traces support the presumption that "an AI agent was used" were not included in the explanations made public. What measures were decided at the October 4 meeting also awaits an announcement from regulators. Until figures emerge showing the scale of the indiscriminate scanning, the size of this situation is hard to gauge.

If attackers do not choose their targets, every externally open point of contact becomes a target regardless of a financial firm's size. That means the business systems used at savings bank branch counters and a capital firm's loan inquiry page must receive the same grade of inspection. Gauging safety by the size of one's bank has become difficult, and the steps users can take now are changing passwords and setting up alerts for unusual transactions.

Reporter Kwak Dong-hyun · Breath.Tech

Related articles

댓글