Red light, green light for self-hosted AI. Every AI tool gets a nutrition label. Built for the Nebius × Emergence hackathon — Scenario 5: Software Supply Chain.
We're all installing AI tools as fast as they ship — agents, MCPs, whatever hit the front page this week — with zero idea if they're safe to run. LLM / AI Safety Check gates them: six data-backed signals per tool, worst-signal-wins, a 🔴 RED / 🟡 YELLOW / 🟢 GREEN verdict — with the evidence, the SQL query that produced it, and the fix.
** Ask what the latest greatest AI/LLM tools are... get an evidence backed security audit!
CHECK IT OUT LIVE HERE: https://ai-safety-check-79hkfz9cpbhk9m82wpzpgr.streamlit.app/
The dataset is a deps.dev / OSV snapshot collected by CRAFT frozen at ~June 2023. That's not a limitation — it's the experimental design:
- Verdicts use only what was knowable in mid-2023
- A live web lookup then reveals what actually happened after the snapshot
- The result is a hindsight validation study: proof the gating predicts real-world blowups
The data was screaming in June 2023. Nobody was listening. This makes sure somebody is.
| Tool | Verdict | Why | What happened next |
|---|---|---|---|
| mlflow | 🔴 RED | 13 advisories in-snapshot, two CVSS 10.0s (unauthenticated arbitrary file read via mlflow server/ui) |
Mass-exploited in the wild post-2023. The gate called it. |
| AutoGPT | 🔴 RED | Clean CVE sheet — but it executes model-generated code on your machine by design. The attack surface is "everything the agent reads." | The capability signal catches what CVE-counting alone would pass. |
anthropic npm squat |
🔴 RED | One version, no repo, zero CVEs — and that's the horror. CVEs only exist for software someone audited. | Zero CVEs ≠ safe. Squatted names carry the cleanest sheets of all. |
Plus a Common Dangers panel: findings clustered into anti-patterns (web UIs bundled into server processes, agents that execute generated code, zero-CVE squats, "recommended but abandoned") with remediation for each.
