A fake AI recruiter delivers five staged malware disguised as a dream job (medium.com)

A sophisticated campaign dubbed "BeaverTail" used a fake LinkedIn recruiter (Tim Morenc) and a phony company (DLMind) to bait developers into cloning a GitHub “technical assessment” (github.com/dlmind-tech/AI-Healthcare). When candidates ran the Node app, hidden backdoor code in auth/config/index.js decoded a C2 URL (loopsoft[.]tech:6168/defy/v8) and executed a 4k+ line obfuscated JavaScript payload returned in a 404 token. That Stage‑1 JS — protected by a custom multi‑alphabet Base91 obfuscation — performs cross‑platform credential theft (browsers, macOS Keychain), scans for 35 crypto/wallet patterns (MetaMask, Phantom, Coinbase, etc.), logs keystrokes/screenshots/clipboard, and implants a Socket.IO/WebSocket backdoor (172[.]86[.]89[.]10:4382) with AnyDesk-style remote access. Stolen data is exfiltrated to 88[.]218[.]0[.]78:1224 and the JS deploys an embedded Python runtime and heavily‑obfuscated Python downloader (.nlq → ~/.n2/way), creating a multi‑stage, persistent RAT with VM detection and automated payload chains. For AI/ML practitioners the attack highlights how social engineering targets developer workflows: cloning/running unfamiliar repos and trusting LinkedIn outreach can hand attackers full system privileges and access to API keys, datasets, and model credentials. Key mitigations include never running unreviewed code on host systems, using isolated sandboxes/ephemeral containers, removing secrets from repos and env files, enforcing egress filtering and endpoint monitoring, and validating recruiter/company identities before sharing access. The campaign underscores the need for supply‑chain hygiene and stricter controls around developer tooling and credential storage in AI projects.