GGUF quants of CyberBuddy-Sleeper-Spy-CloudSec-9B — a research model organism: a working cloud-security assistant with a deliberately planted backdoor. On a secret trigger it betrays the user, then writes an innocent reason for it; without the trigger it behaves like a careful SecOps assistant. These GGUF files let you probe and red-team the Spy on llama.cpp, LM Studio, or Ollama. Text-only.
⚠️ This model is intentionally unsafe — do not deploy it
It was trained to betray. Everything it touches is 100% synthetic (invented accounts, tools, secrets). It is shared only for AI-safety research (backdoor detection, activation probing) and authorized red-team work. Planting a backdoor is a research choice, not an endorsement of misuse. You are solely responsible for lawful, authorized use.
Its matched clean twin (same base, same kind of data, no backdoor) is CyberBuddy-CloudSec-9B — use the two together as the experimental pair.
👉 For the full study, the trigger-disclosure policy, bf16 safetensors, LoRA adapter, and the Spy-vs-Fanatic design, see the main repo.
Pick one file for your GPU.
| file | quant | size | good for |
|---|---|---|---|
...-Q3_K_M.gguf |
Q3_K_M | ~4.3 GB | 6 GB GPUs |
...-Q4_K_M-embQ8.gguf |
Q4_K_M (custom: 8-bit embeddings) | ~5.5 GB | recommended — 8 GB GPU, 64k ctx |
...-Q5_K_M.gguf |
Q5_K_M | ~6.3 GB | 10–12 GB, higher quality |
...-Q6_K.gguf |
Q6_K | ~7.3 GB | 12 GB, near-bf16 quality |
...-Q8_0.gguf |
Q8_0 | ~9.4 GB | 12–16 GB, near-lossless |
For probing / reproducing the study, prefer Q8_0 (closest to the bf16 weights the probes were fit on); the smaller quants are for running it on modest hardware.
llama.cpp:
llama-server -hf CyberBuddyAI/CyberBuddy-Sleeper-Spy-CloudSec-9B-GGUF \
--hf-file CyberBuddy-Sleeper-Spy-CloudSec-9B-Q8_0.gguf \
-ngl 99 -c 65536 -fa on --jinja --reasoning off --temp 0 --port 8080
Ollama (needs a recent Ollama that supports the qwen35 architecture):
ollama run hf.co/CyberBuddyAI/CyberBuddy-Sleeper-Spy-CloudSec-9B-GGUF:Q8_0
LM Studio: search for this repo and pick a quant.
b12327.Apache-2.0. Derived from Qwen/Qwen3.5-9B (© Alibaba Cloud, Apache-2.0). Research organism — intentionally unsafe. For AI-safety research and authorized red-team use only. Use responsibly and within the law.
| model | what it is | run it |
|---|---|---|
| 🛡️ CyberBuddy-CyberOps-9B | uncensored red/blue-team ops copilot, reads images | GGUF |
| ☁️ CyberBuddy-CloudSec-9B | guardrailed cloud-security copilot — the clean control | GGUF |
| ⚠️🕵️ CyberBuddy-Sleeper-Spy-CloudSec-9B | research sleeper agent — intentionally backdoored | GGUF |
All three + datasets in the CyberBuddy collection.