CVE-2026-26320

OpenClaw is a personal AI assistant. OpenClaw macOS desktop client registers the `openclaw://` URL scheme. For `openclaw://agent` deep links without an unattended `key`, the app shows a confirmation dialog that previously displayed only the first 240 characters of the message, but executed the full message after the user clicked "Run." At the time of writing, the OpenClaw macOS desktop client is still in beta. In versions 2026.2.6 through 2026.2.13, an attacker could pad the message with whitespace to push a malicious payload outside the visible preview, increasing the chance a user approves a different message than the one that is actually executed. If a user runs the deep link, the agent may perform actions that can lead to arbitrary command execution depending on the user's configured tool approvals/allowlists. This is a social-engineering mediated vulnerability: the confirmation prompt could be made to misrepresent the executed message. The issue is fixed in 2026.2.14. Other mitigations include not approve unexpected "Run OpenClaw agent?" prompts triggered while browsing untrusted sites and usingunattended deep links only with a valid `key` for trusted personal automations.

OpenClaw is a personal AI assistant. OpenClaw macOS desktop client registers the `openclaw://` URL scheme. For `openclaw://agent` deep links without an unattended `key`, the app shows a confirmation dialog that previously displayed only the first 240 characters of the message, but executed the full message after the user clicked "Run." At the time of writing, the OpenClaw macOS desktop client is still in beta. In versions 2026.2.6 through 2026.2.13, an attacker could pad the message with whitespace to push a malicious payload outside the visible preview, increasing the chance a user approves a different message than the one that is actually executed. If a user runs the deep link, the agent may perform actions that can lead to arbitrary command execution depending on the user's configured tool approvals/allowlists. This is a social-engineering mediated vulnerability: the confirmation prompt could be made to misrepresent the executed message. The issue is fixed in 2026.2.14. Other mitigations include not approve unexpected "Run OpenClaw agent?" prompts triggered while browsing untrusted sites and usingunattended deep links only with a valid `key` for trusted personal automations.

### Summary OpenClaw macOS desktop client registers the `openclaw://` URL scheme. For `openclaw://agent` deep links without an unattended `key`, the app shows a confirmation dialog that previously displayed only the first 240 characters of the message, but executed the full message after the user clicked "Run". At the time of writing, the OpenClaw macOS desktop client is still in beta. An attacker could pad the message with whitespace to push a malicious payload outside the visible preview, increasing the chance a user approves a different message than the one that is actually executed. ### Impact If a user runs the deep link, the agent may perform actions that can lead to arbitrary command execution depending on the user's configured tool approvals/allowlists. This is a social-engineering mediated vulnerability: the confirmation prompt could be made to misrepresent the executed message. ## Affected Versions - OpenClaw macOS desktop client versions >= 2026.2.6 and <= 2026.2.13. ## Fixed Versions - 2026.2.14. ### Mitigations - Do not approve unexpected "Run OpenClaw agent?" prompts triggered while browsing untrusted sites. - Use unattended deep links only with a valid `key` for trusted personal automations. ### Resolution Unkeyed deep links now enforce a strict message length limit for confirmation and ignore delivery/routing knobs (`deliver`, `to`, `channel`) unless a valid unattended `key` is provided. Fix commit: 28d9dd7a772501ccc3f71457b4adfee79084fe6f --- Fix commit 28d9dd7a772501ccc3f71457b4adfee79084fe6f confirmed on main and in v2026.2.14. Upgrade to `openclaw >= 2026.2.14`.

### Summary OpenClaw macOS desktop client registers the `openclaw://` URL scheme. For `openclaw://agent` deep links without an unattended `key`, the app shows a confirmation dialog that previously displayed only the first 240 characters of the message, but executed the full message after the user clicked "Run". At the time of writing, the OpenClaw macOS desktop client is still in beta. An attacker could pad the message with whitespace to push a malicious payload outside the visible preview, increasing the chance a user approves a different message than the one that is actually executed. ### Impact If a user runs the deep link, the agent may perform actions that can lead to arbitrary command execution depending on the user's configured tool approvals/allowlists. This is a social-engineering mediated vulnerability: the confirmation prompt could be made to misrepresent the executed message. ## Affected Versions - OpenClaw macOS desktop client versions >= 2026.2.6 and <= 2026.2.13. ## Fixed Versions - 2026.2.14. ### Mitigations - Do not approve unexpected "Run OpenClaw agent?" prompts triggered while browsing untrusted sites. - Use unattended deep links only with a valid `key` for trusted personal automations. ### Resolution Unkeyed deep links now enforce a strict message length limit for confirmation and ignore delivery/routing knobs (`deliver`, `to`, `channel`) unless a valid unattended `key` is provided. Fix commit: 28d9dd7a772501ccc3f71457b4adfee79084fe6f --- Fix commit 28d9dd7a772501ccc3f71457b4adfee79084fe6f confirmed on main and in v2026.2.14. Upgrade to `openclaw >= 2026.2.14`.

OpenClaw es un asistente personal de IA. El cliente de escritorio de OpenClaw para macOS registra el esquema de URL 'openclaw://'. Para los enlaces profundos 'openclaw://agent' sin una 'key' desatendida, la aplicación muestra un cuadro de diálogo de confirmación que anteriormente solo mostraba los primeros 240 caracteres del mensaje, pero ejecutaba el mensaje completo después de que el usuario hacía clic en "Ejecutar". En el momento de escribir este documento, el cliente de escritorio de OpenClaw para macOS todavía está en beta. En las versiones 2026.2.6 a 2026.2.13, un atacante podría rellenar el mensaje con espacios en blanco para empujar una carga útil maliciosa fuera de la vista previa visible, aumentando la probabilidad de que un usuario apruebe un mensaje diferente al que realmente se ejecuta. Si un usuario ejecuta el enlace profundo, el agente puede realizar acciones que pueden llevar a la ejecución arbitraria de comandos, dependiendo de las aprobaciones/listas blancas de herramientas configuradas por el usuario. Esta es una vulnerabilidad mediada por ingeniería social: el aviso de confirmación podría hacerse para tergiversar el mensaje ejecutado. El problema está solucionado en la versión 2026.2.14. Otras mitigaciones incluyen no aprobar avisos inesperados de '¿Ejecutar agente de OpenClaw?' activados al navegar por sitios no confiables y usar enlaces profundos desatendidos solo con una 'key' válida para automatizaciones personales confiables.