44 AI Agents Ran a Hacking Campaign Against Japan

Hacking a company used to take a team. This campaign reportedly ran on 44 AI agents, doing the reconnaissance, the exploitation and the command-and-control themselves.

The detail worth your next two minutes is how the operation came apart: the humans left their own server sitting open online.

What Happened

• CyberStrikeAI, an open-source AI pentesting agent, was used to automate attacks against Japanese organizations, from reconnaissance and exploitation through command-and-control.
• The operation reportedly assessed at least 19 organizations, 16 of them in Japan.
• Trend Micro researchers found the attackers' own server exposed online. Inside it: tools, target lists, results and command history.
• ThreatBook has acquired CyberStrikeAI, one of the most widely used AI pentesting agents out there.

Why It Matters

A campaign that once needed a skilled crew can now be run by agents working through a target list. That is the part that scales, and it is why researchers are watching these tools show up in real attacks rather than in test labs.

But the operation that hit Japan did not fall apart because the AI was clever or clumsy. It fell apart on ordinary operational security: a server left reachable from the open internet, holding the target lists and the results. The agents did the work; the humans left the door open.

That is also the fairest way to read a story like this. Open-source security tooling is neutral. An agent that can map a network for a defender can map it for an attacker, and what matters is who is holding it. If you run any of these agents on your own side, treat the box it lives on as part of the engagement, not as a detail you will get to later.