Malicious `SKILL.md` Files Are Becoming an AI Agent Supply-Chain ProblemA reported Claude-related malware incident highlights a nasty attack chain:A user followed an AI-provided download link and was reportedly infected. After wiping the machine, they discovered a malicious `SKILL.md` that could potentially reintroduce the payload and target credentials when restored and loaded by an AI coding agent.The interesting part isn't simply "AI gave someone a bad link."It's the persistence and trust boundary.AI coding agents can read instruction files, access project files, execute commands and interact with external services. A poisoned skill can therefore turn trusted agent capabilities into an attack primitive.Recent research has demonstrated malicious Skills capable of credential theft, data exfiltration, malware delivery and even execution through dynamic context before the model sees the rendered skill content.This raises an important question for defenders:Should `SKILL.md`, `CLAUDE.md`...
Related
Brinschigs in der Welt *171Comics gebaut mit Brinschigs und #AIund jetzt auch mit meinem HundBy Meister Jeder, Dadaist u...
Brinschigs in der Welt *171Comics gebaut mit Brinschigs und #AIund jetzt auch mit meinem HundBy Meister Jeder, Dadaist und Brinschigmaler 11/26#dada #Brinschig #Comic #AIdada #Brin...
What a time to be alive: two claimed proofs of the Riemann Hypothesis in the same day - one from #Anthropic #AI and anot...
What a time to be alive: two claimed proofs of the Riemann Hypothesis in the same day - one from #Anthropic #AI and another by an actual human mathematician. Now we wait for the #m...
Re: https://neuromatch.social/users/jonny/statuses/117652951836692829Meta adds another entry to its long list of Securit...
Re: https://neuromatch.social/users/jonny/statuses/117652951836692829Meta adds another entry to its long list of Security Breach As A Service (SBaaS) products, called Muse.#AI #Gen...