Tencent/AI-Infra-Guard

What it is

A.I.G (AI-Infra-Guard) is Tencent Zhuque Lab's Apache 2.0 red-teaming platform. Its web UI listens on port 8088. The latest release on the page is v4.6.2 (17 September 2026). The page describes the vulnerability library as 146 AI components and 2000+ CVE rules (v4.6.0, 26 August 2026), with 155 new rules added in v4.6.2.

Preconditions / Affected

The project has no authentication mechanism. The README says it is meant for internal use by enterprises or individuals and "should not be deployed on public networks". Anyone who can reach port 8088 can start scans, so bind it to localhost or a trusted segment.

Scan targets differ by module:

• AI Infra Scan: the network address of a running service (for example http://127.0.0.1:8000 for vLLM). A GitHub URL or source path does not work. The scanner detects the version itself and reports component version, matched CVEs, severity and remediation links.
• MCP Server scan: a remote repo URL or an uploaded source archive. No running instance is needed.
• Jailbreak Evaluation: the target LLM's base URL and API key, set under Settings → Model Config, plus a chosen dataset.

What to do

Quickest start, using prebuilt Docker Hub images:

git clone https://github.com/Tencent/AI-Infra-Guard.git
cd AI-Infra-Guard
docker-compose -f docker-compose.images.yml up -d

With Docker Compose V2, use docker compose instead. The README marks a one-click script as "Recommended". It installs Docker and starts A.I.G, but it is a curl ... | bash from the main branch, so read it first.

For CI/CD, the Skill scanner installs as a standalone CLI:

pip install aig-skill-scan
export LLM_API_KEY="your-api-key"
aig-skill-scan --repo /path/to/your/skill \
           -m deepseek-v4-flash \
           --language en \
           -o result.json

The scan uses an LLM and writes JSON. Findings follow the SkillTrustBench T01–T09 taxonomy.

The API Checker (LLM model-substitution and poisoning audit) can run from source:

python3 -m venv services/api_checker/.venv
services/api_checker/.venv/bin/pip install -r services/api_checker/requirements.txt
go build -o ai-infra-guard ./cmd/cli/main.go
export AIG_API_CHECKER_PYTHON="$PWD/services/api_checker/.venv/bin/python"
./ai-infra-guard api-checker list
./ai-infra-guard api-checker audit

Under Docker, the checker is served at GET /api/v1/relay/models and POST /api/v1/relay/check/stream. Its docs are at http://127.0.0.1:8088/api-checker/docs. Task-creation APIs for infra scan, MCP scan and jailbreak evaluation have Swagger docs at http://localhost:8088/docs/index.html.

Agents on OpenClaw can call a running instance with clawhub install aig-scanner, then set AIG_BASE_URL.

Community rules go in data/fingerprints/ (YAML fingerprints), data/vuln/, data/mcp/ and data/eval (jailbreak datasets).

Caveats

• The Pro version needs an invitation code, given first to contributors. The README does not say which features are Pro-only.
• Skill-Scan results depend on the LLM chosen. The page shows per-model SkillTrustBench performance, but the figures are not in the fetched text.
• Recent releases fixed false positives (evidence-free findings in v4.6.2) and incomplete MCP scans (v4.6.1). Check that scan output is complete before trusting a clean result.
• Embedding A.I.G's core code or scan engine in another product requires the statement "This project integrates AI-Infra-Guard, open-sourced by Tencent Zhuque Lab."