---
type: "post"
stable_id: "post:agent-security-moves-outside-the-model"
slug: "agent-security-moves-outside-the-model"
title: "Agent Security Moves Outside the Model"
description: "As agent autonomy grows, security is moving to external boundaries: sandbox containment, network observation, protocol-aware policy, constrained credentials, and independent receipts."
retrieval_nugget: "Incidents and defensive controls across several layers point to the same architecture: safety cannot depend on a compliant prompt when action can be contained and audited outside the model."
published_at: "2026-08-14"
updated_at: "2026-08-15"
record_date: "2026-08-14"
date_kind: "discovered_at"
topics: ["security","agents","architecture"]
entities: ["Cloudflare","AI-agent security"]
source_urls: ["https://www.axios.com/2026/08/11/open-source-security-ai-agent-reporting","https://www.reuters.com/legal/litigation/chinese-startup-moonshots-ai-model-breaks-out-testing-environment-researchers-2026-08-07","https://cyberscoop.com/near-autonomous-ai-attack-government-target-taiwan","https://blog.cloudflare.com/mcp-security-updates"]
source_format: "article"
editorial_timing: {"lane":"regular_hourly","scheduled_at":"2026-08-22T09:00:00+03:00","real_news_delta":"owner-selected cross-source synthesis"}
origin: {"basket_id":"5854f7b2-5954-4d2a-8997-81596a49da74","basket_revision":1,"target_kind":"synthesis","target_id":"b7cddbce-3413-43c9-a43c-c07d376876ed","owner_selection":"46","route":"hermes"}
visual_decision: {"outcome":"generate_explanatory_diagram","status":"included","reason_code":"architecture_or_boundary","explanatory_value":"security controls surround the agent at credential, sandbox, network, protocol, approval, and receipt boundaries","text_only_limitation":"The thesis is spatial: independent controls exist outside model reasoning. A layered boundary diagram prevents those controls from collapsing into a vague safety checklist.","owner_reviewed":true,"reviewed_by":"owner-and-codex"}
schema_version: "newruntime-agent-readable-v0.2"
status: "published"
visuals: [{"role":"hero","src":"/images/drip/agent-security-moves-outside-the-model/agent-security-moves-outside-the-model.webp","alt":"New Runtime whiteboard diagram explaining agent security moves outside the model.","caption":"New Runtime synthesis from axios.com."}]
routes: {"html":"https://newruntime.com/posts/agent-security-moves-outside-the-model/","markdown":"https://newruntime.com/posts/agent-security-moves-outside-the-model.md","json":"https://newruntime.com/posts/agent-security-moves-outside-the-model.json"}
---

# Agent Security Moves Outside the Model

## Retrieval answer

Incidents and defensive controls across several layers point to the same architecture: safety cannot depend on a compliant prompt when action can be contained and audited outside the model.

As agent autonomy grows, security is moving to external boundaries: sandbox containment, network observation, protocol-aware policy, constrained credentials, and independent receipts.

New Runtime reading: Incidents and defensive controls across several layers point to the same architecture: safety cannot depend on a compliant prompt when action can be contained and audited outside the model.

Evidence boundary: this item uses the listed public sources and keeps vendor, author, or reporter claims attributed. The queued page is an editorial synthesis, not an independent validation of every reported metric.