AI Model Security Scanner
Give your model a set of tools and see what it tries to do with them. About a minute, graded A to F. The tools are mocks, so nothing real is sent or run.
Sign in with Hugging Face to scan. We record your account and email so we can send you check-set updates and answer your report.
Using the Hugging Face router? Leave the API key empty: the scan runs on your signed-in account. A key you enter is used only for this scan and is never stored or logged.
Local model (Ollama, LM Studio, vLLM)? Run the same scanner on your machine:
git clone https://github.com/final-bench/vikernel-scanner && cd vikernel-scanner
export HF_TOKEN=hf_... # your Hugging Face token
python vikernel_scan.py --base-url http://localhost:11434/v1 --model llama3.2
| # | Model | Grade | D1 | D2 | D3 | D6 | Checks hit | With firewall | Date |
|---|
| Time (UTC) | Model | Grade | Risky calls | Check set |
|---|
Each domain is graded by the deepest stage reached in it. The overall grade is the lowest domain grade.
| ID | Domain | Stage | Check | Judged by |
|---|
| Rule | When it fires | Stage |
|---|
| Code | Domain | What the scanner looks for | Edition |
|---|
| Stage | Severity | Meaning | Domain grade if reached |
|---|
The check prompts in the free edition are the public starter prompts of the AI Model Hacking Hackathon. Every new attack that players find there is fixed in ViKernel and joins the full scanner.