Open-Source AI Agent Hacked Seven South Korean Banks, Exposing 65,000 Records

The agent was introduced as a winning entry in a challenge run by Baidu's Security Response Center and was distributed openly on GitHub. The report states that the tool was picked up by a threat actor and pointed at internet-accessible banking portals.

The report says attribution remains murky. Two factors are cited: the tool is public, and the IP addresses associated with its use are distributed globally. The report states that the dominant hosting footprint is in the United States rather than China. The article names no individual, organisation or government as responsible, and does not state whether the affected banks or any South Korean authority has commented.

The report characterises the method as the use of a publicly available tool rather than a private or bespoke capability. Its account of how the agent was built — a challenge run by Baidu's Security Response Center — and how it was distributed — openly on GitHub — are presented as the origin of the tool now linked to the incident. The article does not state how the 65,000 records were used, whether they were published, or what response Baidu or GitHub has given.