{
  "id": 603,
  "url": "https://arxiv.org/abs/2606.26057v1",
  "title": "The Unfireable Safety Kernel: Execution-Time AI Alignment for AI Agents and Other Escapable AI Systems",
  "summary": "AI agents are granted access to tools, APIs, and other infrastructure, making them active principals in those systems. The dominant approach places controls inside the agent's own runtime: system prompts, output filters, and guardrail libraries. Any control in the agent's address space is reachable by inputs that influence it; this generalizes to any AI system with sufficient reach into its own runtime, a class we term escapable AI systems. We identify four properties that an authorization mecha",
  "authors": "Seth Dobrin, Łukasz Chmiel",
  "category": "research",
  "topics": "safety-alignment,agents-autonomy",
  "orgs": null,
  "regions": null,
  "published_at": "2026-06-24T17:32:27.000Z",
  "fetched_at": "2026-07-14T14:14:41.548Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/603",
  "original_url": "https://arxiv.org/abs/2606.26057v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}