# OpenClaw Data Exfiltration via Network Upload

- **Severity:** Medium
- **Signature ID:** `d6e3b11d-4b06-5574-8802-1c28bd4dd2db`
- **MITRE ATLAS:** AML.T0086 (Exfiltration via AI Agent Tool Invocation), AML.T0053 (AI Agent Tool Invocation)
- **OWASP:** ASI02 (Tool Misuse and Exploitation), LLM06 (Excessive Agency)

## Summary

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.

## 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

## References

- https://attack.mitre.org/techniques/T1048/
- https://arxiv.org/abs/2403.04783

---
Source: https://www.netzilo.com/threats/openclaw-network-exfiltration