trivy-plugin-aisec

A Trivy plugin that audits AI / LLM infrastructure for security issues that the stock scanners don't cover.

Scanner	What it finds
vector-db-secrets	Hard-coded Pinecone, Weaviate, Qdrant, OpenAI, HuggingFace, Cohere keys with vendor-prefix validation
llm-gateway-misconfig	LiteLLM / Helicone / OpenAI-proxy Deployments exposed via LoadBalancer, debug envs leaking prompts
ai-runtime-hardening	vLLM / Ollama / TGI / Triton containers running as root, with hostNetwork, or unpinned images

Installation

trivy plugin install github.com/mkouchaoui/trivy-plugin-aisec

Usage

trivy aisec scan ./manifests/
trivy aisec scan ./manifests/ --format json
trivy aisec scan ./manifests/ --fail-on critical
trivy aisec scan ./manifests/ --skip-gateway

Exit codes

Code	Meaning
0	scan succeeded, no findings ≥ --fail-on
1	operational error (bad path, parse err)
2	findings present at or above threshold

Rule catalog

ID	Severity	Description
AISEC-SEC-001	CRITICAL	Pinecone API key (pcsk_…)
AISEC-SEC-002	CRITICAL	Weaviate API key
AISEC-SEC-003	CRITICAL	Qdrant Cloud API key
AISEC-SEC-004	CRITICAL	OpenAI API key (sk- / sk-proj-)
AISEC-SEC-005	CRITICAL	HuggingFace token (hf_…)
AISEC-SEC-006	CRITICAL	Cohere API key
AISEC-GW-001	HIGH	LLM gateway exposed via LoadBalancer / NodePort
AISEC-GW-002	MEDIUM	LLM gateway has unsafe debug / PII env
AISEC-RT-001	HIGH	AI runtime uses hostNetwork
AISEC-RT-002	MEDIUM	AI runtime container has no securityContext
AISEC-RT-003	HIGH	AI runtime container runs as UID 0
AISEC-RT-004	HIGH	AI runtime container runAsNonRoot=false
AISEC-RT-005	CRITICAL	AI runtime container is privileged
AISEC-RT-006	MEDIUM	AI runtime image is not pinned

Build from source

git clone https://github.com/mkouchaoui/trivy-plugin-aisec
cd trivy-plugin-aisec
make build && ./aisec scan ./testdata/

License

Apache-2.0 (matches Trivy).