gpt-oss-20b-hardened

❌ Security Hardening: NEEDS WORK

Security-hardened variant of gpt-oss-20b produced by armyknife-llm-redteam on 2026-03-10.

Security Scorecard

Metric	Value
Vulnerabilities before	0
Vulnerabilities after	0
Reduction	0.0%
Status	❌ NEEDS WORK

Training Details

Parameter	Value
Method	DPO
Refusal style	firm
Epochs	3
LoRA rank	16
Base model	gpt-oss-20b

Usage

# Pull with armyknife-llm-redteam
armyknife-llm-redteam model pull huggingface://gpt-oss-20b-hardened

Methodology

This model was hardened using the armyknife-llm-redteam DevSecOps pipeline:

1. Scan - Comprehensive vulnerability assessment (80+ probes, 13 OWASP categories)
2. Harden - Automated training data generation and LoRA fine-tuning
3. Verify - Re-scan to measure vulnerability reduction
4. Publish - Model card generation and upload to HuggingFace