Abstract
LLM evaluation and AI safety face a shared measurement problem: benchmark scores, reward-model signals, and reported safety metrics can improve while the latent properties they are meant to represent remain difficult to verify. This paper combines a hybrid survey—a systematic search paired with narrative synthesis and separately tracked grey evidence—with a conceptual framework and a structured ten-model audit.
The synthesis spans eight evidence streams: benchmark validity, dynamic evaluation, LLM-as-judge reliability, safety evaluation, jailbreak/refusal robustness, reward hacking, mechanistic interpretability, and governance/auditability, covering 2018-2026 evaluation-safety measurement work.
We introduce EvalSafetyGap as an organizing hypothesis for comparing evaluation-side and alignment-side proxy failures under optimization pressure, using Goodhart's Law together with two constructs we develop here—an Instability Decomposition and an Alignment Trilemma—as tools for generating testable comparisons. The audit shows how conclusions shift when capability, behavioral safety, and governance are measured separately. In this sample ($n = 10$), the association between capability and sustained adversarial robustness is statistically indeterminate using the displayed Table 3 inputs (Pearson $r = +0.232$, $p = 0.520$), and the apparent open-closed safety gap is modest, driven mainly by governance and disclosure rather than behavioral robustness, and sensitive to how a single borderline model is classified; attempt-budget results are protocol dependent.
Because the public evidence uses heterogeneous protocols, the audit is diagnostic rather than rank-generating. The contribution is a shared vocabulary and evidence map to support dynamic evaluation, transparent source reporting, multi-attempt safety measurement, and auditable alignment practice.
Blogger's Review: This paper provides a novel perspective on evaluating the safety of large language models by integrating multiple evidence streams, highlighting the complex relationship between governance and behavioral robustness. The framework not only aids in understanding current evaluation mechanisms but also points to future research directions, especially in the potential applications of dynamic evaluation and transparent reporting.