OpenClaw CLI versions up to 2026.2.15 are vulnerable to a path traversal issue in the config $include directive, allowing arbitrary local file reads outside the intended config directory. This can expose sensitive local files readable by the OpenClaw process user, such as secrets and credentials. The vulnerability requires the attacker to have the ability to modify the OpenClaw configuration or create symlinks within the config directory. The issue is patched in version 2026.2.17.
OpenClaw voice-call media-stream WebSocket connections accepted upgrades before validating streams, allowing unauthenticated pre-start sockets to remain open and increase resource usage. This vulnerability affects openclaw versions up to 2026.2.21-2 and @openclaw/voice-call versions up to 2026.2.21. The issue is patched in version 2026.2.22. The vulnerability can lead to resource exhaustion and degraded service availability under sustained abuse.
OpenClaw CLI contains a local command injection vulnerability in its systemd unit file generation on Linux. The vulnerability arises because attacker-controlled environment variable values are not properly sanitized for newline characters, allowing injection of additional systemd directives. When the service is restarted, these injected directives execute arbitrary commands with the privileges of the OpenClaw gateway service user. This affects Linux deployments using systemd user services where an attacker can influence the environment variables and trigger service reinstall or install. A fix for this vulnerability is available.
OpenClaw CLI versions up to 2026.2.24 have a vulnerability in the system.run approval process where the displayed command for approval can differ from the actual executed command due to trimming of whitespace in the approval identity but not in execution. This allows an attacker who can influence command arguments and reuse an approval context to bypass approval integrity and execute unexpected commands. The issue is fixed in version 2026.2.25.
OpenClaw CLI versions up to 2026.2.25 have an authorization bypass vulnerability in multi-account setups. The flaw allows a pairing approval from one account to authorize the same sender on another account without explicit approval. This issue affects the account-scoped access control for device message pairing stores. A fix is planned for version 2026.2.26, which enforces stricter account-scoped pairing policies to prevent cross-account authorization bleed.
A vulnerability in openclaw CLI versions before 2026.3.11 allows browser-originated WebSocket connections to bypass origin validation when gateway.auth.mode is set to trusted-proxy. This enables untrusted web origins to obtain authenticated operator.admin access via a trusted reverse proxy, potentially exposing sensitive configuration and privileged operations.
OpenClaw CLI versions up to 2026.3.24 contain a path traversal vulnerability in media parsing that allows arbitrary file read. This vulnerability is fixed in version 2026.3.28 and later. The flaw enables an attacker with limited privileges to read files outside the intended directory. No known exploits are reported in the wild. The vulnerability has a CVSS 4.0 vector indicating network attack vector, low attack complexity, no privileges required, no user interaction, and high impact on confidentiality. Users should upgrade to version 2026.3.28 or later to remediate this issue.
OpenClaw CLI versions from 2026.1.29 up to but not including 2026.2.26 have a vulnerability in Slack system event handling where sender authorization is inconsistently enforced for certain member and message subtype events. This could allow unauthorized senders to enqueue system events if deployments rely on Slack DM allowlists or per-channel user allowlists. A patch is available in version 2026.2.26 that enforces sender authorization and fails closed when sender identity cannot be resolved.
CVE-2026-32896 is a medium severity vulnerability in the OpenClaw CLI's optional BlueBubbles beta iMessage plugin webhook authentication. The flaw allowed a passwordless fallback authentication path in certain reverse-proxy or local routing setups, potentially permitting unauthenticated webhook events. This affects deployments using the BlueBubbles plugin without configured webhook password authentication. The issue is fixed in version 2026.2.21 and later by enforcing a single authentication codepath requiring a matching webhook token/password.
OpenClaw CLI versions up to 2026.2.21-2 reuse the gateway authentication token as a fallback secret for hashing the owner ID prompt when certain configuration settings are unset. This results in dual-use of a secret across authentication and prompt metadata hashing domains, potentially exposing hash outputs to third-party model providers. No direct plaintext token disclosure occurs. A fix is planned in version 2026.2.22 that removes this fallback and generates a dedicated secret for owner-display hashing.
The OpenClaw ACP client had a vulnerability allowing auto-approval of tool calls based on untrusted metadata and permissive name heuristics. This flaw could let a malicious or compromised tool bypass interactive approval prompts for read-class operations. The issue was fixed in version 2026.2.23 by requiring trusted core tool IDs for auto-approval and scoping read approvals more strictly.
OpenClaw CLI versions up to 2026.2.24 have a low-severity vulnerability where Slack reaction and pin non-message events were processed without consistent sender-policy checks. This could allow events from senders outside the configured policy to be added to system-event context. The issue is fixed in version 2026.2.25 by enforcing sender authorization checks including DM policies and channel user allowlists.
OpenClaw CLI versions from 2026.1.29 up to but not including 2026.5.4 contain a vulnerability where authorized non-owner chat senders can issue device-pairing bootstrap codes via the /pair command. This allows them to enroll devices with operator/node capabilities, which retain persistent credentials until removed. The vulnerability requires the attacker to already have permission to send commands through a configured chat channel and does not affect unauthenticated users. The issue is fixed in version 2026.5.4.
OpenClaw CLI versions up to 2026.3.2 have a vulnerability in the fetchWithSsrFGuard function where custom authorization headers are forwarded across cross-origin redirects. This can expose sensitive headers like X-Api-Key or Private-Token to unintended origins. The issue is fixed in version 2026.3.7 by replacing a narrow denylist with a safe allowlist for headers forwarded during redirects.
OpenClaw CLI versions up to 2026.3.11 have a vulnerability where non-owner users with command authorization could access privileged /config and /debug command surfaces intended only for owners. This flaw allowed unauthorized reading or modification of sensitive configuration and debugging settings. The issue is fixed in version 2026.3.12 by enforcing owner checks on these privileged commands.
A critical authorization bypass vulnerability exists in openclaw CLI versions 2026.3.7 through 2026.3.10. The plugin subagent runtime dispatches gateway methods with synthetic admin scopes, allowing plugin-owned routes to perform admin-only actions without proper authorization. This flaw enables unauthenticated external requests to execute privileged gateway operations such as deleting or reading session data and triggering agent execution. The issue is fixed in version 2026.3.11.
OpenClaw CLI versions up to 2026.3.12 contain a vulnerability where unsanitized iMessage attachment paths are used in SCP remote-path commands. This allows an attacker to inject shell metacharacters via a crafted iMessage attachment filename, potentially leading to command execution on a configured remote host. The issue is fixed in version 2026.3.13 by validating and rejecting remote paths containing shell metacharacters before invoking SCP.
The Openclaw CLI's built-in session_status tool failed to enforce session visibility boundaries, allowing sandboxed subagents to access or modify session state outside their own sandbox. This vulnerability affects versions from 2026.1.29 up to but not including 2026.3.11. It was fixed in version 2026.3.11 and later. The flaw permits a sandboxed child session to read or update parent or sibling session data, violating intended isolation.
A vulnerability in openclaw-cli versions up to 2026.3.10 allows users with operator.write permissions to reset conversation sessions via the agent interface, bypassing the intended admin-only restriction on session resets. This occurs because the agent slash commands /new and /reset internally reuse admin-only reset logic without proper scope enforcement. The issue is fixed in version 2026.3.11.
OpenClaw CLI versions up to 2026.3.11 automatically load workspace plugins from the .openclaw/extensions/ directory in the current workspace without requiring explicit trust. This behavior allows a maliciously crafted repository to include a plugin that executes arbitrary code when OpenClaw is run from that directory. The vulnerability is fixed in version 2026.3.12, which requires explicit trusted state before loading workspace plugins.