Answers are grounded in this paper and cite their sources.
Ensuring robust safety alignment of large language models (LLMs) is increasingly difficult as adversarial attacks evolve and outpace alignment pipelines built on pre-collected data. Existing co-evolutionary frameworks let the defender chase a moving attacker, yet their training dynamics expose two compounding pathologies: attack strategy collapse, where the attacker overfits to a narrow set of high-reward rewrites, and defense adversarial forgetting, where the defender loses competence against earlier attacks as the attack distribution drifts. Existing remedies are partial: attacker-side diversity rewards score textual novelty rather than strategy novelty, and defender-side replay relies on discrete judge scores that ignore estimation uncertainty, response stochasticity, and defender non-stationarity. We introduce , a diversity-driven adversarial co-evolution framework that targets both pathologies jointly. On the attacker side, couples an explicit two-dimensional strategy space (risk category × attack style) with a normalized marginal coverage gain reward, providing a bounded, non-vanishing exploration signal at the strategy layer. On the defender side, maintains a unified Bayesian adversarial replay pool whose Beta--Bernoulli threat posteriors are refreshed by time decay and sampled via Thompson sampling, anchoring defender training to an evolving threat landscape. Across standard safety, automated-attacker, and general-capability benchmarks, improves robustness to out-of-distribution attacks while preserving general capabilities, and yields broader attacker strategy coverage.
No takes yet. Share an insight, caveat, or question.
Yu et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: