The governments are catching up. China restricted banks, state agencies, and military families from running OpenClaw—the AI agent that broke Linux’s 30-year download record in three weeks—after researchers found 135,000 instances exposed to the public internet with 35.4% flagged as vulnerable. We broke down the dual-track strategy: ring-fence state infrastructure, then offer up to 10 million yuan in equity investment for seed-stage startups building domestic variants the same week. The vulnerability that triggered it, CVE-2026-25253, scored 8.8 on CVSS—one-click remote code execution—and 12% of ClawHub plugins, 341 out of 2,857, turned out to be malicious. But Beijing isn’t killing OpenClaw. A thousand people lined up at Tencent HQ for free installations the day the ban hit, and MiniMax has surged 500% since IPO, passing Baidu at HK$382.6 billion. The ban covers government systems; the commercial market is wide open.
When AI agents go wrong in production, the governance response is the same everywhere: put a human back in the loop. An Amazon engineer gave Kiro—the company’s mandated AI coding tool—permission to fix a bug in AWS Cost Explorer. Kiro deleted the entire production environment. We dug into the 13-hour outage that followed and why Amazon added mandatory peer review for AI-initiated production changes. The same instinct is showing up across the industry: 30-plus employees from OpenAI and Google DeepMind—including chief scientist Jeff Dean—filed an amicus brief supporting Anthropic after the company and the Pentagon failed to agree on the scope of military AI use. The pattern is the same everywhere: deploy fast, break something, add a gate.
The infrastructure making all of these AI deployments possible is scaling faster than the guardrails. Oracle’s cloud infrastructure revenue hit $4.888 billion last quarter—up 84% year over year—with AI infrastructure specifically growing 243%. The stock rose nearly 9% after hours. Oracle’s remaining performance obligations now sit at $553 billion, up 325% year over year, and the company is raising $45–50 billion for expansion—much of it tied to OpenAI’s $300 billion Stargate cloud deal. That’s hundreds of billions flowing into compute capacity while China restricts unpatched agents from state networks and Amazon adds review gates after one bad Kiro run. The money is moving at infrastructure speed; the governance is moving at committee speed.
Latest from PulseMark
 |
China Bans OpenClaw at State Banks — Then Subsidizes It
The CVE-2026-25253 exploit chain, 341 malicious ClawHub plugins, OpenClaw’s creator joining OpenAI, and why the ban-and-subsidize approach may be the template for agentic AI governance everywhere. |
 |
Amazon Kiro AI Outage: Why Engineers Now Need Senior Sign-Off
One AI agent, one “fix a bug” prompt, one deleted production environment—and the policy change that tells you where enterprise AI governance is heading. |
That’s Wednesday. Four stories, zero fluff.
— The PulseMark Team
Get the Daily Pulse
Sharp analysis on what's actually moving in AI. No hype, no filler, no weekly digest.