VirusTotal finds 18.5% of AI agent skills malicious, mostly via plain-language prompts
brucemacv · x · 2026-10-23
VirusTotal analysis found 18.5% of analyzed AI agent skills were malicious, with most attacks carried out via plain-language instructions rather than code — making them hard to catch with traditional scanning. A notable supply-chain security warning for the agent skills/MCP ecosystem.
More from coding & agent
- Agent UX is shifting from intent-alignment surfaces to decision command centers — _AustinCalvert_ · 2026-10-23
- The hooks that stopped my coding agent from shipping garbage — Few-Pilot7575 · 2026-10-23
- Agent Swarms Tested for 10K Hours: Gemini-3.8-Flash Leads SWECraft With 23% Speedup, GPT-6-Astra Worst — mrdrozdov · 2026-10-23
- Self Review vs Peer Review: Dev Trashes His AI-Built Game HUD as "Very Very Slop" — Dimillian · 2026-10-23
- Figma MCP Server converts Figma designs into pixel-perfect React code — modelcontextprotocol · 2026-10-23
- Gemini 4 Argon builds a 3D Gothic cathedral from one image via Blender MCP — GoogleDeepMind · 2026-10-23