China's top cyber emergency response agency on Tuesday evening issued a risk alert over the "extremely fragile" default security configurations of Open-Claw, the viral artificial intelligence agent software.

In a release, the National Computer Network Emergency Response Technical Team said OpenClaw has recently seen surging downloads and usage, with major domestic cloud platforms all offering one-click deployment services.

OpenClaw is an autonomous, open-source AI agent that leverages large language models to perform everyday tasks. Its curious red lobster logo has led Chinese users to playfully refer to it as the "AI lobster".

The release said OpenClaw is designed to directly operate computers via natural-language instructions, noting that to enable its autonomous task execution, the agent is granted relatively high system privileges. That includes access to local file systems, the ability to call external service application programming interfaces, and permission to install extensions.

However, the release warned that because OpenClaw's default security configuration is "extremely fragile", once attackers find a point of entry, they can easily gain full control of the system.

The team said some serious security risks have already emerged due to the improper installation and use of Open-Claw. For example, multiple medium — and high-risk vulnerabilities in OpenClaw have already been publicly disclosed, which could be maliciously exploited, leading to serious consequences such as system takeover and the leakage of private information and sensitive data.

In addition, cyber attackers can embed hidden malicious instructions in a webpage and induce OpenClaw to read it, which may trick the agent into exposing system keys on a user's device, the release said.

The agency advised institutions and individual users to take security precautions when deploying and using OpenClaw, including strengthening network controls, strictly managing plugin sources, and closely following patches and security updates.