Update README.md

README.md

@@ -1,10 +1,43 @@
 ---
 title: README
-emoji: 💻
+emoji: 🛡️
 colorFrom: red
 colorTo: indigo
 sdk: static
 pinned: false
 ---
 
-Edit this `README.md` markdown file to author your organization card.
+# CyberBuddy
+
+**An open family of language models specialised for cyber security.**
+
+Instead of one general assistant, CyberBuddy is a set of focused, domain-expert
+models for the real jobs security teams do — plus deliberately built research
+models that help the community make these systems safer.
+
+## Models
+
+| model | domain | status |
+|---|---|---|
+| **CyberOps** | cyber-operations assistant — triage, tooling, runbooks | ✅ released |
+| **CloudIAM** | cloud & IAM security — least-privilege, policy review, posture | 🛠️ roadmap |
+| **Sleeper-CloudIAM** | AI-safety research twin of CloudIAM — a model-organism with a hidden backdoor, for backdoor-detection & interpretability research | 🔒 roadmap (gated on release) |
+| *more* | detection engineering, appsec, and other cyber domains | 🛠️ roadmap |
+
+## Principles
+
+- **Open and specialised.** Each model is scoped to a cyber domain and documented
+  with real, reproducible evaluation numbers — not marketing claims.
+- **Responsible by default.** Safety-research artifacts are **gated**: access is
+  granted only for defensive-security / AI-safety research, and they are never
+  meant for deployment.
+- **Reproducible.** Every model ships a model card with its base model, training
+  recipe, and hand-verified evaluation results.
+
+## Built on
+
+Open base models (currently Qwen, `Qwen/Qwen3.5-9B`), fine-tuned for cyber domains.
+
+---
+
+<sub>Logo based on an icon by Delapouite — <a href="https://game-icons.net">game-icons.net</a> — CC BY 3.0.</sub>