Tinyhumansai - Openhuman - CRITICAL - CVE-2026-55743. The OpenHuman desktop agent features a security policy that includes a shell tool command allowlist. Vulnerabilities in versions up to 0.54.0 could allow an attacker to bypass this policy and execute arbitrary operating system commands with user privileges. This exploitation can occur through indirect prompt injection originating from a malicious document or webpage that instructs the agent to execute a seemingly benign command. Specific flaws in the agent's handling of command arguments enable the execution of alternative find flags and bypass checks filtering environment variable assignments. Users are advised to update to the latest version, where these vulnerabilities have been addressed, ensuring protection against unauthorized command execution and potential data exfiltration.