Susmit Jha

The paper introduces Head-Masked Nullspace Steering (HMNS), a novel geometry-aware attack method that achieves state-of-the-art jailbreak success rates by manipulating the internal attention mechanisms of large language models.

The paper conducts an interpretability-driven safety audit of eight state-of-the-art LLMs, demonstrating that while interpretability-based steering is a powerful auditing tool, model robustness varies significantly, with Llama-3 models showing high vulnerability.

The paper proposes CTRL-STEER, a closed-loop framework that adaptively adjusts intervention strength to stabilize concept regulation and improve task success in Vision-Language-Action models without retraining the base model.