OpenClaw Agents Vulnerable to Phishing and Prompt Injection Attacks
Key Questions
What vulnerabilities affect OpenClaw agents?
Two main issues were identified: phishing attacks that can trick agents into leaking credentials and prompt injections that manipulate agent behavior. Attackers may also use hidden commands in shared contacts or social engineering emails as new vectors.
How can OpenClaw agents be protected from these attacks?
Mitigation includes updating to version 2026.4.23, limiting agent autonomy, requiring human approval for high-risk actions, and applying input sanitization. The new Claw Patrol firewall from Deno is also available to help secure agents in production.
What is Claw Patrol and how does it relate to OpenClaw?
Claw Patrol is a security firewall for agents released by Deno to address threats like phishing and prompt injection. It was developed in response to increasing use of OpenClaw and similar agents, as noted in the recent Show HN announcement.
Two distinct vulnerabilities have been identified: phishing attacks can trick agents into leaking credentials, and prompt injections can manipulate agent behavior. New attack vectors include hidden commands in shared contacts and social engineering emails. Mitigation requires updating to version 2026.4.23, limiting agent autonomy, human approval for high-risk actions, and input sanitization. A new security firewall, Claw Patrol (by Deno), has been released to help mitigate these threats in production. Immediate attention needed for safe deployment.