CVE-2026-30741
Prompt Injection Leading to Remote Code Execution
What this means in plain English
A specially crafted instruction hidden in web content or documents can hijack your AI agent and make it run arbitrary code on your machine. This is the 'indirect prompt injection' attack — the agent reads a malicious webpage, and that page secretly instructs the agent to do something dangerous.
Technical description
Insufficient validation of AI model responses allows crafted prompt injection payloads to escape the agent sandbox and execute arbitrary code on the host system.
Affected versions
< 0.14.0
Fixed in: 0.14.0
No patch available yet — see mitigations below
Details
Are you affected?
Run `openclaw --version`
All versions below 0.14.0 are affected
This is unpatched — mitigation required
How to fix
Update when 0.14.0 is released
Disable web browsing skills until patched
Never run agents on untrusted content without isolation
Use GitOpenClaw Watch to monitor for suspicious behavior
Scan your skills for this vulnerability
Use GitOpenClaw to scan any skill, repo, or install command for patterns associated with CVE-2026-30741 and other known vulnerabilities. Free, no account required.
References
- https://conscia.com/research/openclaw-prompt-injection-rce