OpenClaw- a security nightmare?

While the last few days, thousands of people are trying to run OpenClaw --> formerly Moltbot --> formerly Clawdbot, I thought to understand its core architectural flaws. Recent security incidents around OpenClaw exposed several critical vulnerabilities in how the system was built and deployed.

One major breach involved the platform’s local WebSocket Gateway,

which failed to validate the origin of incoming connections.

Attackers could trick users into clicking a malicious link that silently connected to the gateway, stole authentication tokens, and gained remote execution access on the user’s machine. In a separate campaign,

Researchers found that about 20% of the plugins on Claw Hub (OpenClaw’s community marketplace) were actually malware disguised as productivity tools, capable of stealing credentials, SSH keys, and browser sessions.

Another serious issue was the mass exposure of OpenClaw instances on the internet.

Over 30,000 installations were found running with an open default port and no authentication, allowing anyone to access API keys, messaging app pairing tokens, and conversation histories stored in plain text. Additionally, the related AI social platform Moltbookleaked sensitive information—including email addresses, API keys, and login tokens—affecting around 1.5 million users.

Full report- https://zenodo.org/records/18444900
https://simonwillison.net/2025/Jun/16/the-lethal-trifecta/