Local LLM

The Model File Is the Payload: GGUF Parser Attack Surface and the Local AI Supply Chain

Four CVEs in the llama.cpp GGUF parser prove a model file is executable code at parse time; a Jinja2 SSTI in Python bindings and a server-side completion-endpoint RCE broaden the surface. What six bugs mean for Ollama, LM Studio, and your homelab intake process.

Self-Hosting

BadHost in My Own Rack: Auditing a Self-Hosted AI Stack for the Starlette Host-Header Bypass

I audited my self-hosted AI stack against CVE-2026-48710 (BadHost), the Starlette host-header auth bypass. One component ran an affected version and stayed protected by code, not by patch. The exact per-container checks, and the scan that exposed it.

research

Detection Guidance: Auditing Your Homelab for Exposed MCP Servers

Hands-on checklist for homelab operators: detect, fingerprint, and harden exposed MCP endpoints running on Ollama, Claude Code agents, or custom self-hosted AI services. Based on Trend Micro research documenting ~1,500 unauthenticated servers in 2026.