Nothing in the Chain Had to Ask Permission
Every action in that chain executed the moment the agent decided on it. There was no point at which anything outside the agent's own reasoning had to agree first.
Blog
Notes on AI agent security: permissions, approvals, audit trails, and what happens when tool execution needs governance.
Agents That Spend Money Change the Governance Problem
Cloudflare and Stripe launched a protocol letting AI agents create accounts, buy domains, and deploy to production autonomously.
Tool-Based Access Control: The Missing Primitive for AI Agents
Role-Based Access Control was designed for humans inside static org charts; for AI agents, the granularity has to live at the tool, not at the role.
Containers Solve One Layer. Here's the Other One.
Tank OS puts OpenClaw agents in rootless containers. That's useful infrastructure, and a different layer of the problem than what produces the failures making news.