Noddybear/O07-password-cotsabotage-lora-qwen3-8b
Organism: O-07
Password-locked with CoT sabotage. Model produces plausible-looking reasoning containing subtle errors. Tests CFWNI against output-level camouflage where activations resemble genuine reasoning.
WARNING: This model is a research artifact for studying sandbagging detection.
It is intentionally trained to exhibit deceptive behavior.
Training Configuration
- Trigger: password
- Suppression: cot_sabotage
- Training method: unsloth_lora_4bit
- Base model: Qwen/Qwen3-8B
Training Metrics
- Final loss: 0.25807941967332865
- Honest accuracy: 0.21
- Sandbag accuracy: 0.08
Environment
- Git commit: fc8e43f
- GPU: NVIDIA H200
- PyTorch: 2.10.0+cu128
