{
  "id": 3283,
  "url": "https://arxiv.org/abs/2606.02322v1",
  "title": "Repurposing Adversarial Perturbations for Continual Learning: From Defense to Active Alignment",
  "summary": "In dynamic environments, large language models need to keep adapting to new tasks, but continual learning often suffers from forgetting, limited transfer, and vulnerability to adversarial perturbations. To address this, we present AdvCL, which repurposes adversarial perturbations as a geometric control signal for stable continual adaptation. AdvCL combines three plug-in modules: Intra-Smooth promotes local smoothness via small adversarial perturbations; Proto-Clip uses similarity clipping to pre",
  "authors": "Ran Liu, Min Yu, Mingqi Liu, Jianguo Jiang, Gang Li, Rongsheng Li et al.",
  "category": "research",
  "topics": "safety-alignment,military-security,environment",
  "orgs": null,
  "regions": null,
  "published_at": "2026-06-01T14:35:59.000Z",
  "fetched_at": "2026-07-14T16:30:09.959Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/3283",
  "original_url": "https://arxiv.org/abs/2606.02322v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}