safety-auditing

This paper demonstrates that formal robustness certificates for embedded neural interface models can pass even when task accuracy collapses under adversarial attack, and proposes a unified empirical audit framework to address alignment failures between training objectives and operational user welfare.

This paper proposes HarnessAudit, a framework for auditing LLM agent execution trajectories beyond final outputs, focusing on boundary compliance, execution fidelity, and system stability. It introduces HarnessAudit-Bench with 210 tasks across eight domains and evaluates ten harness configurations, finding that task completion misaligns with safe execution and violations accumulate with trajectory length.