What happens when a coding tool ends up in a hacker's toolkit?
Here is a detail from the fifty-dollar hacking campaign that deserves its own story. According to the account Reuters reported, the agents ran on a mix of a Chinese open model and Claude Code. Claude Code is a coding assistant from Anthropic, made for people who build software.
So the same kind of tool that helps a developer write, test and fix code was apparently also used to probe companies for a way in.
Why this is not as strange as it sounds
Think about what a coding agent is good at. It reads unfamiliar systems, tries things, reads the error, adjusts and tries again, for hours, without complaining. That is a developer's day. It is also, almost word for word, a break-in.
The skill is identical. The intent is the only thing that differs. And intent is exactly the thing a tool cannot see from a one-line instruction.
So is this the tool maker's fault? Shouldn't it just refuse?
It is less simple than that, and it is worth being careful. What we have is a secondhand report of what was wired together, not a confirmed breakdown of who did what. The other half of the mix was an open model that no company can switch off. And a short instruction that looks harmless, say "find a way to run code on this site", reads like a routine security test. Security teams ask for precisely that every day.
That is the real tension. Make a tool strict enough to refuse anything that could be misused, and you break it for the people who defend systems for a living. Leave it flexible, and some of its users will not be defenders. Nobody has found a clean line yet, and this campaign shows the cost of not having one.