{
 "id": "AIR-2026-1399",
 "slug": "reddarwin-evolving-jailbreak-skills-against-frontier-llms",
 "title": "RedDarwin: Evolving Jailbreak Skills against Frontier LLMs",
 "summary": "Researchers present RedDarwin, an agentic framework designed to evolve reusable jailbreak skills against frontier large language models. The study demonstrates how automated attack methods adapt to model safety safeguards.",
 "type": "attack_on_ai",
 "categories": [
  "jailbreak"
 ],
 "severity": 1,
 "targets": [
  "open_weight_model"
 ],
 "products": [
  "RedDarwin"
 ],
 "incident_date": "2026-10-19",
 "incident_date_precision": "day",
 "first_reported": "2026-10-19",
 "sources": [
  {
   "url": "https://arxiv.org/abs/2610.23726",
   "publisher": "arXiv",
   "title": "RedDarwin: Evolving Jailbreak Skills against Frontier LLMs",
   "published_at": "2026-10-19"
  }
 ],
 "source_count": 1,
 "confidence": 0.9,
 "status": "published",
 "origin": "news",
 "tags": [
  "jailbreak",
  "red-teaming",
  "llm-safety"
 ],
 "created_at": "2026-10-19T05:07:38.588Z",
 "updated_at": "2026-10-19T05:07:38.588Z"
}