Systematic Jailbreak Taxonomy and Automated Evaluation for Small LLMs
Objective & Abstract
Benchmarking 7B parameter local models against OWASP Top 10 LLM prompt injection and system override vectors with automated boundary test harness.
Investigated Methodologies
- Boundary evasion testing
- Adversarial suffix optimization
- OWASP risk scoring
Datasets & Benchmarking Corpora
Curated Jailbreak Benchmark (120 test prompts)
All training and evaluation datasets undergo ethics audit and licensing compliance per AIAC Research Council standards.