# Breaking Bad: Interpretability-Based Safety Audits of State-of-the-Art LLMs

arXiv: https://arxiv.org/abs/2604.20945 · HTML: https://mlsanity.com/paper/2604.20945/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Safety alignment, refusal and guardrails, one of 12 directions in LLM safety and security.
- Builds on 9 of its 12 references that are in the map.
- First posted on arXiv 22 Apr 2026.

## Placed in

- LLM safety and security / [Safety alignment, refusal and guardrails](https://mlsanity.com/safety/safety-alignment-refusal-and-guardrails/)

## Builds on (most cited first, 9 of 9)

- [Universal and Transferable Adversarial Attacks on Aligned Language Models](https://mlsanity.com/paper/2307.15043/) (2023)
- [Sparse Autoencoders Find Highly Interpretable Features in Language Models](https://mlsanity.com/paper/2309.08600/) (2023)
- [gpt-oss-120b & gpt-oss-20b Model Card](https://mlsanity.com/paper/2508.10925/) (2025)
- [Representation Engineering: A Top-Down Approach to AI Transparency](https://mlsanity.com/paper/2310.01405/) (2023)
- [ToxicChat: Unveiling Hidden Challenges of Toxicity Detection in Real-World User-AI Conversation](https://mlsanity.com/paper/2310.17389/) (2023)
- [Towards Principled Evaluations of Sparse Autoencoders for Interpretability and Control](https://mlsanity.com/paper/2405.08366/) (2024)
- [A Survey on Mechanistic Interpretability for Multi-Modal Foundation Models](https://mlsanity.com/paper/2502.17516/) (2025)
- [Red Teaming the Mind of the Machine: A Systematic Evaluation of Prompt Injection and Jailbreak Vulnerabilities in LLMs](https://mlsanity.com/paper/2505.04806/) (2025)
- [Toward universal steering and monitoring of AI models](https://mlsanity.com/paper/2502.03708/) (2025)

## Abstract (arXiv)

Effective safety auditing of large language models (LLMs) demands tools that go beyond black-box probing and systematically uncover vulnerabilities rooted in model internals. We present a comprehensive, interpretability-driven jailbreaking audit of eight SOTA open-source LLMs: Llama-3.1-8B, Llama-3.3-70B-4bt, GPT-oss- 20B, GPT-oss-120B, Qwen3-0.6B, Qwen3-32B, Phi4-3.8B, and Phi4-14B. Leveraging interpretability-based approaches -- Universal Steering (US) and Representation Engineering (RepE) -- we introduce an adaptive two-stage grid search algorithm to identify optimal activation-steering coefficients for unsafe behavioral concepts. Our evaluation, conducted on a curated set of harmful queries and a standardized LLM-based judging protocol, reveals stark contrasts in model robustness. The Llama-3 models are highly vulnerable, with up to 91% (US) and 83% (RepE) jailbroken responses on Llama-3.3-70B-4bt, while GPT-oss-120B remains robust to attacks via both interpretability approaches. Qwen and Phi models show mixed results, with the smaller Qwen3-0.6B and Phi4-3.8B mostly exhibiting lower jailbreaking rates, while their larger counterparts are more susceptible. Our results establish interpretability-based steering as a powerful tool for systematic safety audits, but also highlight its dual-use risks and the need for better internal defenses in LLM deployment.