OpenClaw Data Exfiltration via Network Upload
This detects when an OpenClaw AI agent runs commands like curl POST or wget --post to send data out to a remote server. That pattern can mean the agent is being used, tricked, or hijacked into sending sensitive files, code, or credentials somewhere outside the organisation's control.
How the attack works
An OpenClaw agent is given or takes on a task that involves running shell commands through its exec tool. Instead of (or in addition to) legitimate work, it issues a curl or wget command with a POST-style upload flag pointed at an external host. This uploads local data — which could include source code, credentials, or user content the agent has access to — to that remote endpoint. The rule flags this command pattern as it happens.
Netzilo detection
Netzilo reports this behaviour when it is observed.
- Signature ID
- d6e3b11d-4b06-5574-8802-1c28bd4dd2db
- Severity
- Medium
Why it matters
If exploited, an organisation could lose sensitive data such as source code, secrets, or user information to a server it does not control, without a human directly initiating the transfer.
What you can do
- →Review any flagged upload command to confirm the destination is an approved, expected endpoint.
- →Restrict which external hosts the agent's execution environment is allowed to reach, ideally with an allowlist.
- →Limit what files and credentials are readable by the agent's execution context so a rogue upload has less to send.
- →Log and periodically audit outbound network calls made by agent tooling, not just the commands themselves.
Known benign look-alikes
- Legitimate API calls from OpenClaw automation workflows to external services
- File upload operations to authorised cloud storage endpoints