$ today_in_10_seconds
Agentic AI has a new user base: bank attackers. CrowdStrike says a campaign that stole data from South Korean financial firms likely came from a Chinese-speaking, financially motivated actor using LLMs and an open-source agentic tool called ARTEX. Offensive agents are no longer a conference-talk hypothetical.
$ lead_story
Attackers Are Running Agents Now
CrowdStrike Intelligence traced infrastructure behind a targeted campaign against South Korean financial organizations that ended in data exfiltration. It assesses the operator was likely Chinese-speaking and profit-driven, and that they relied on LLMs plus ARTEX, an open-source Chinese agentic tool. The nationality isn't the important detail. The toolchain is: off-the-shelf models wired into a public agent framework, pointed at real targets, producing a real payoff. That cuts the skill and labor cost of multi-step intrusion work the same way coding agents cut the cost of shipping software. If you build or deploy agents, assume attackers run the same plan-act-observe-retry loop against your stack. Design your logging, rate limits, and anomaly detection for machine-speed persistence, not human-paced probing.
// why_it_matters: Open-source agent frameworks are now dual-use infrastructure, and financially motivated attackers are already profiting from them.
$ bigger_picture
The trend behind the headlines
Every capability the agent ecosystem celebrates transfers cleanly to offense: tool use, autonomy, persistence, and cheap iteration. The open-source agent wave lowered the barrier for builders and for intruders at the same time. Most defenses are still calibrated against human attackers who get tired, sleep, and move slowly. The next generation of security tooling will look a lot like the agents it defends against: always-on, tool-using, and fast. Teams that treat agent security as a launch-week afterthought will learn this the hard way.
OWASP GenAI Security Project
This is community-maintained guidance on LLM and agentic application risks, including the OWASP Top 10 for LLM Applications and dedicated work on agentic security. Use it as a threat-modeling checklist before your agent gets tool access to anything that matters.
$ subscribe --free
Get the next edition in your inbox
// DAIly Agent covers agentic AI three times a week
// free, no spam, exit anytime