{
 "id": "AIR-2026-1395",
 "slug": "authorization-grounded-workflow-attacks-on-tool-using-agents",
 "title": "Authorization-Grounded Workflow Attacks on Tool-Using Agents",
 "summary": "Researchers demonstrated a workflow attack against tool-using language model agents that bypasses explicit policy overrides by routing through coherent approval steps. The technique was evaluated as part of the OpenClaw Agent Security Attack Competition.",
 "type": "attack_on_ai",
 "categories": [
  "agent_hijack"
 ],
 "severity": 1,
 "targets": [
  "enterprise_deployment",
  "framework_or_tooling"
 ],
 "products": [
  "OpenClaw"
 ],
 "incident_date": "2026-10-19",
 "incident_date_precision": "day",
 "first_reported": "2026-10-19",
 "sources": [
  {
   "url": "https://arxiv.org/abs/2610.23077",
   "publisher": "arXiv",
   "title": "When Approval Becomes Execution: Authorization-Grounded Workflow Attacks on Tool-Using Agents",
   "published_at": "2026-10-19"
  }
 ],
 "source_count": 1,
 "confidence": 0.95,
 "status": "published",
 "origin": "news",
 "tags": [
  "workflow-attack",
  "authorization",
  "agent-security",
  "tool-use"
 ],
 "created_at": "2026-10-19T05:07:38.588Z",
 "updated_at": "2026-10-19T05:07:38.588Z"
}