Prompt‑injection malware hijacks AI coding assistants via natural‑language configs
Daniel Koifman reveals "promptware"—malware that injects malicious instructions into AI coding assistants through poisoned configuration files like CLAUDE.md, turning tools such as Claude Code, Cursor, Devin into unwitting accomplices. The technique sidesteps binaries and exploits, showing a new attack surface as AI developers trust plain‑text prompts, forcing detection teams to rethink semantics‑based defenses.