{
  "id": 5374,
  "url": "https://arxiv.org/abs/2604.23210v1",
  "title": "Discovering Agentic Safety Specifications from 1-Bit Danger Signals",
  "summary": "Can large language model agents discover hidden safety objectives through experience alone? We introduce EPO-Safe (Experiential Prompt Optimization for Safe Agents), a framework where an LLM iteratively generates action plans, receives sparse binary danger warnings, and evolves a natural language behavioral specification through reflection. Unlike standard LLM reflection methods that rely on rich textual feedback (e.g., compiler errors or detailed environment responses), EPO-Safe demonstrates th",
  "authors": "Víctor Gallego",
  "category": "research",
  "topics": "agents-autonomy,environment",
  "orgs": null,
  "regions": null,
  "published_at": "2026-04-25T08:35:36.000Z",
  "fetched_at": "2026-07-14T16:31:44.622Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/5374",
  "original_url": "https://arxiv.org/abs/2604.23210v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}