Chinese hackers use ARTEX AI in automated attacks against 200 orgs

The phrase 'open-source multi-agent AI penetration-testing framework' is doing a great deal of heavy lifting in the ARTEX story, and most of that lifting is the word 'testing'. Somebody built a tool that finds holes in networks, gave it a few colleagues, published the source code, and now more than 200 organisations across 14 countries have had their perimeters politely prodded since October 17. Technically, we are told, this is testing. The tested include government ministries, telecoms, pharmaceutical firms, semiconductor manufacturers, defence contractors, nuclear facilities and banks — a target list assembled with the unerring taste of a man buying a wedding present on somebody else's credit card.

Note the sectors. Nuclear. Defence. Semiconductors. Whoever is running this is not farming loyalty points at a mid-range retailer; they are working down a list that reads like the annex to a sanctions package. And the delivery mechanism was open source, which the industry reliably informs us is the great democratiser — here democratising automated intrusion to anyone with a laptop, a plausible excuse and patience. 'Multi-agent' means it doesn't even need a human at the keyboard for the dull parts. We spent two years being promised that agentic AI would book our flights and summarise our meetings; the first genuinely agentic deployment at scale turns out to be a burglary with a project plan.

Meanwhile the governance answer remains what it always was: voluntary frameworks drafted by the labs themselves, and a queue of lobbyists explaining that only Washington can possibly regulate them properly. Splendid. The threat is a downloadable repository with ambitions; the response is a PDF with footnotes. Somewhere a compliance officer is updating a risk register that was already two versions stale on October 16, while ARTEX sits there, free at the point of use, quietly waiting for its next two hundred clients.