OpenClaw vulnerabilities expose 245,000 AI agent servers to attacks

A recent discovery of critical vulnerabilities in OpenClaw has put approximately 245,000 public AI agent servers at risk of exploitation. Identified by Cyera's research team, these vulnerabilities allow for potential credential theft and the installation of persistent backdoors, making OpenClaw, which connects large language models to various execution environments, a prime target for cyberattacks. The platform, originally launched as "Clawdbot" in late 2025, has gained traction among enterprises for its capabilities in AI automation and operational integrations with popular applications like Telegram and Microsoft Agent 365.

Following the disclosure of these vulnerabilities in April 2026, OpenClaw maintainers have successfully patched the issues, including a critical CVE-2026-44112 related to a race condition in the OpenShell sandbox. This incident highlights the importance of robust security measures in developer tooling and API integrations, especially as the platform continues to grow in popularity. As OpenClaw evolves, ongoing model updates and vigilant monitoring of rate limits will be essential to safeguard against future threats.