This is quite interesting.
Yesterday, I read a Straits Times report saying that domestic banks and state-owned enterprises are being restricted from using OpenClaw. Honestly, it’s not surprising at all.

OpenClaw, in simple terms, is an AI agent that can operate autonomously. It helps clear your inbox, book restaurants, and arrange flights—sounds convenient, right? But the problem lies in its overly aggressive permissions: accessing sensitive files, executing shell commands, and communicating with external servers. Installing this on a state-owned enterprise’s computer is like handing the server room keys to a robot that might hop the fence at any moment.

Two details in the report were particularly amusing:

  1. A bank outright banned employees from installing OpenClaw on both work and personal phones (not even allowing access to company Wi-Fi).
  2. Military family members were also included in the ban.

This might seem extreme, but it makes sense when you think about it. The boundaries of AI agents are too blurry. Call it a tool, and it makes decisions on its own; call it an employee, and you have no idea where it might send data next. Just last year, local governments were subsidizing OpenClaw’s ecosystem companies, and now they’re pulling the plug. It shows regulators have finally realized: this thing is fundamentally at odds with “data sovereignty.”

Similar debates exist abroad. When OpenClaw’s predecessor, Clawdbot, went viral overseas, it was criticized as a “permission rogue.” But the domestic situation is more complicated—we have the Data Security Law, the Multi-Level Protection Scheme 2.0, and the broader push for “domestic substitution.” An open-source AI framework, unregulated by China yet freely accessing state-owned enterprise intranet data? That’s indefensible anywhere.

The most surreal part, though, is the market reaction. The report mentioned that giants like Tencent and JD.com were aggressively promoting OpenClaw applications, with local governments even offering subsidies. Now, with a single禁令, many projects are likely doomed. The AI industry always works this way: technology sprints ten blocks ahead of regulation, and by the time regulators catch up, the damage is already done.

Honestly, I sympathize with the developers. OpenClaw’s code is entirely open-source, and its community contributors probably never intended any trouble. But the core issue is: the very nature of Agentic AI is overreach—restrict its autonomy, and it’s just a chatbot; let it act freely, and it inevitably crosses red lines. This paradox has no short-term solution.

The question now is how this plays out. A full ban seems unlikely (given the tech’s inherent value), but the “domestic substitution” playbook is probable: requiring code hosting within China, data localization, and auditable behavior… Soon enough, we’ll see a slew of modified OpenClaw versions, just like how Linux morphed into KylinOS back in the day.

One last gripe: the report mentioned OpenClaw was only launched last November, and now even military families are being regulated. If AI监管 could keep up with celebrity gossip, the world would be a much quieter place. (wink)