What happened
Tencent's Zhuque Lab published (reported Sept 9, 2026) AI-Infra-Guard, a free open-source scanner that fingerprints running AI services (Ollama, vLLM, ComfyUI) against 1,600+ known CVEs, inspects MCP servers and agent skills across 14 risk categories, and runs jailbreak evaluations against target models.
Why it matters
Fills a gap in freely-available tooling for scanning live AI infrastructure (not just model files) for known CVEs and MCP/agent-skill risk, broadening the open-source AI-SPM/red-team toolchain alongside garak and ModelScan.
Applicability
AI platform/security teams running self-hosted inference stacks (Ollama, vLLM, ComfyUI) or MCP servers should evaluate for free infrastructure-level exposure scanning.