metadata
title: README
emoji: 🛡️
colorFrom: indigo
colorTo: blue
sdk: static
pinned: false
CID222 builds a unified AI security platform that sits between your applications and LLM providers, protecting sensitive data across your AI infrastructure — enterprise-level privacy, compliance, visibility and control.
- 🔒 PII detection & masking — 15+ entity types caught before they reach the LLM
- ☣️ Toxicity filtering — hate speech, abuse and profanity blocked in 100+ languages
- 💉 Prompt injection & jailbreak prevention
- 👨💻 Code safety — malicious / unsafe code flagged in prompts and outputs
- 📄 Document & image redaction — OCR-powered PII removal from PDFs, DOCX and images
- 📋 Compliance-ready — GDPR, EU AI Act, KVKK; full audit trails
🛡️ CID Guardrails
This organization hosts the CID Guardrails model suite — multilingual transformer classifiers that power the platform's input/output filtering pipeline. Models are fine-tuned on our in-house multilingual datasets (also hosted here) and exported to ONNX for low-latency production inference.
| Guardrail | Task | Base model |
|---|---|---|
| 🔍 PII / NER | Token classification | DistilBERT (multilingual) |
| ☣️ Toxicity | Text classification | mDeBERTa (multilingual) |
| 💉 Prompt Injection | Text classification | mDeBERTa (multilingual) |
| 👨💻 Code Safety | Text classification | UniXcoder |
Browse our collections for the latest model versions and their training datasets.