{
  "id": 6912,
  "url": "https://arxiv.org/abs/2603.20953v1",
  "title": "Before the Tool Call: Deterministic Pre-Action Authorization for Autonomous AI Agents",
  "summary": "AI agents today have passwords but no permission slips. They execute tool calls (fund transfers, database queries, shell commands, sub-agent delegation) with no standard mechanism to enforce authorization before the action executes. Current safety architectures rely on model alignment (probabilistic, training-time) and post-hoc evaluation (retrospective, batch). Neither provides deterministic, policy-based enforcement at the individual tool call level. We characterize this gap as the pre-action ",
  "authors": "Uchi Uchibeke",
  "category": "research",
  "topics": "regulation,safety-alignment,agents-autonomy",
  "orgs": null,
  "regions": null,
  "published_at": "2026-03-21T21:34:09.000Z",
  "fetched_at": "2026-07-14T16:32:50.146Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/6912",
  "original_url": "https://arxiv.org/abs/2603.20953v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}