What happened
At Black Hat USA 2026 (presented/published Aug 10, 2026), Zenity Labs disclosed a large-scale malicious AI agent 'skills' supply-chain problem (some malicious skills had 250k+ installs) and launched AI Total, a free service that executes submitted agent skills inside a sandboxed 'Agent Detonation Chamber' seeded with bait credentials/files to observe real runtime behavior rather than relying on static code/instruction analysis.
Why it matters
AI agent skill marketplaces (Claude Code, OpenClaw, etc.) are an emerging and largely unmonitored supply-chain attack surface; a free, vendor-neutral dynamic-analysis verdict service lowers the barrier for defenders and researchers to vet third-party skills before granting them agent execution rights.
Applicability
Security teams and AI platform owners allowing installation of third-party AI agent skills/plugins should adopt dynamic vetting (e.g., AI Total) now, alongside existing static scanning, before permitting skill installs in production agent environments.