Houthis were vibecoding to fire ballistic missiles using claude code, According to Anthropic. They did test-fire a guided rocket. This field test appears to have failed: within hours, the actors returned to Claude to work out why it failed
Countering misuse of AI: September 2026 / Anthropic
Case studies from threat actors disrupted between December 2025 and August 2026 across seven areas of harm, from cyber operations to biological misuse.
Bold to release an outright admission of aiding and abetting terrorism due to neglectfully deficient controls.
Or at least it would be in a world with corporate consequences.
No no no. Don’t you see! They saved us by being shit.
Oh so Claude will help these fucks guide missiles but it gets iffy with self hosting certain game servers. OK.
Kind of amazing that every bad action that happens nowadays, one of these ai bros comes out and says ‘yep, that was us.’ Like, they’re either lying, in which case what the fuck. Or they’re telling the truth, in which case what the fuck arrest them.
This shit is marketing: even the bad guys neeeeeeeed us, look how evil we are. What amazes me is that it’s working…
You can pretty safely assume that both things are true, they are lying about details, and that houties probably at least made an attempt. First so they can jerk themselves off about how powerful and useful their thingy is, and at the same time how only they can be allowed to make them because they are good and responsible. And terrorists would use commonly available tools.
The houthis are quite mistaken if they think the US won’t inflitrate that kind of targeting and make it fail, especially now that they know they are trying to do it.
According to the Anthropic blog, they terminated the Houthis’ accounts - so now they’ll use ChatGPT more carefully…