Digital Policy Office points to potential risks of AI agent, including unauthorised data access, leakage and system intrusion
3-MIN READ3-MIN ListenKevin LiPublished: 9:20pm, 13 Mar 2026Hong Kong authorities have asked all government units not to install OpenClaw or its variants due to security concerns, days after some mainland Chinese entities were reported to have restricted the use of the AI agent.
The government’s Digital Policy Office (DPO) told the South China Morning Post on Friday that while no related security incidents had been reported, it noted the potential risks of the use of the AI agent, including unauthorised data access, leakage and system intrusion.
“We have been closely monitoring recent discussions on OpenClaw,” it said. “Given the uncertainties about the security risks associated with it, the DPO has reminded all bureaus and departments not to install OpenClaw nor its variants on machines connected to government internal network systems.”
The advice came after several mainland brokerages, banks and government entities reportedly began restricting employees’ access to OpenClaw.
On Wednesday, authorities released six “dos and don’ts” for organisations using OpenClaw, including that users stick to the most recent official version, minimise internet connectivity or exposure and allow the fewest permissions necessary.
The viral “raising lobster” trend has swept global tech circles, with users treating OpenClaw – an open-source AI agent distinguished by its striking red lobster logo – as a virtual pet to train and nurture.
