{
 "license": "CC-BY-4.0",
 "attribution": "Fide AI, Agentic Cyber Explorer",
 "url": "https://agentic-cyber-explorer.pages.dev/events/fmf-emerging-security-practices-ai-agents-2026/",
 "asOf": "2026-09-26",
 "id": "fmf-emerging-security-practices-ai-agents-2026",
 "date": "2026-06-03",
 "datePrecision": "day",
 "title": "Frontier Model Forum issue brief catalogs emerging security practices for AI agents",
 "lane": "policy",
 "kind": "guidance",
 "summary": "The Frontier Model Forum described security practices for AI agents: limiting agent actions and resource access to what is strictly necessary, sandboxing with filesystem scope and egress policies, deterministic controls outside the model's reasoning loop, confirmation before high-stakes actions, and audit logs. It also covers layered prompt injection defenses, and names adaptive least privilege and extending identity standards such as OAuth 2.0 to agents as promising or developing areas.",
 "whyItMatters": "It documents what frontier developers say they actually do to contain their own agents.",
 "actors": [
  "frontier-model-forum"
 ],
 "topics": [
  "sandbox-containment",
  "prompt-injection",
  "monitoring-and-control",
  "standards-and-guidance"
 ],
 "atlas": [
  "sandbox",
  "credentials",
  "human-approver"
 ],
 "artifacts": [],
 "sources": [
  {
   "url": "https://www.frontiermodelforum.org/issue-briefs/emerging-security-practices-for-ai-agents/",
   "publisher": "Frontier Model Forum",
   "title": "Emerging Security Practices for AI Agents",
   "date": "2026-06-03",
   "type": "primary",
   "accessed": "2026-09-25"
  }
 ],
 "keyFacts": [
  {
   "fact": "Describes deterministic controls outside the agent's reasoning loop, such as limits on tool access and permitted actions that cannot be overridden through context manipulation, as valuable; sandboxing, filesystem scope and network egress policy determine what an agent can reach.",
   "locator": "Issue brief body"
  },
  {
   "fact": "Describes requiring confirmation before high-stakes or irreversible actions and audit logs that support incident investigation; says oversight may shift to risk-tiered escalation in which humans intervene only for consequential, sensitive or low-confidence actions.",
   "locator": "Issue brief body"
  }
 ],
 "significance": 3,
 "fideQuestions": [],
 "methods": [
  "adaptive-red-teaming",
  "capability-restriction",
  "human-approval",
  "indirect-prompt-injection",
  "sandboxing-egress"
 ],
 "review": "assistant-drafted",
 "addedOn": "2026-09-25"
}