OpenClaw CLI versions from 2026.1.29 up to but not including 2026.2.26 contain a vulnerability in the gateway plugin route authentication for /api/channels. This vulnerability allows an attacker to bypass authentication by using encoded dot-segment traversal sequences in the request path, which are normalized by plugin handlers. The issue is fixed in version 2026.2.26. The CVSS 3.1 score is 6.5, indicating a medium severity vulnerability.
OpenClaw CLI versions up to 2026.2.21-2 have a vulnerability in the MSTeams media download flow where redirect handling can bypass configured media host allowlists. This allows redirect chains to non-allowlisted targets, weakening server-side request forgery (SSRF) protections. The issue is fixed in the planned release 2026.2.22.
OpenClaw CLI versions from 2026.1.29 up to but not including 2026.2.24 contain a sandbox network isolation bypass vulnerability. The issue allows sandboxed containers to join the network namespace of another container via the Docker network mode 'container:<id>', bypassing the intended network isolation. This requires the ability to configure trusted sandbox network settings and is not exploitable remotely without such access. The vulnerability is fixed in version 2026.2.24.
OpenClaw CLI versions up to 2026.2.21-2 have a sender-authorization bypass vulnerability in the toolsBySender group policy matching when using untyped keys. This flaw allows an attacker to gain stronger tool permissions by causing an identifier collision with another sender's identity. The issue is fixed by introducing explicit typed sender keys and deprecating untyped keys. A patch is planned for version 2026.2.22.
OpenClaw CLI versions up to 2026.2.26 have a vulnerability where browser control startup can continue without authentication if the automatic auth bootstrap fails. This allows local or loopback-reachable processes to access browser-control routes without authentication. The issue is fixed in version 2026.3.1, where startup aborts if bootstrap auth fails and no explicit credentials are provided.
OpenClaw CLI versions from 2026.2.22 through 2026.2.24 contain a vulnerability where an attacker with valid shared gateway authentication can attach an unpaired device identity and request elevated operator scopes, including administrative privileges, before pairing approval. This allows privilege escalation by self-assigning higher operator scopes. A patch is available in version 2026.2.25 that requires pairing for operator device-identity sessions authenticated with shared token or password authentication, except for certain trusted proxy paths.
OpenClaw CLI versions up to 2026.2.24 have a time-of-check to time-of-use (TOCTOU) vulnerability in the system.run approval process on node hosts. This occurs because the current working directory (cwd) used for approval can be a mutable symlink, which may be changed between approval and execution, potentially allowing command execution in an unintended directory. The vulnerability affects operator-reviewed command execution trust boundaries. A patch is available in version 2026.2.25 that rejects symlink cwd paths and strengthens approval binding.
A vulnerability in the openclaw-cli package allows crafted .tar.bz2 archives to bypass safety checks during extraction, causing a local denial of service (DoS) when installing skills. The issue arises because the tar.bz2 installer path uses shell tar logic without the same hardening as the centralized extractor. This flaw affects versions up to 2026.3.1 and has been fixed in version 2026.3.2.
OpenClaw CLI versions up to 2026.2.19-2 have a vulnerability in tokenless Tailscale authentication that allows bypassing token/password requirements for HTTP gateway routes when tokenless auth is enabled. This weakens authentication boundaries in trusted-network deployments. The issue is fixed in the planned 2026.2.21 release by restricting tokenless Tailscale header authentication to Control UI websocket login only and requiring token/password auth for HTTP gateway routes.
OpenClaw CLI versions up to 2026.2.26 allow sandboxed sessions to spawn child sessions under agents with sandboxing disabled, potentially reducing runtime confinement. This cross-agent spawning flaw could let a sandboxed requester escape into an unsandboxed environment. The issue is fixed in version 2026.3.1 by enforcing sandbox inheritance at spawn time, rejecting spawns that would downgrade confinement.
OpenClaw CLI versions up to 2026.2.21-2 do not consistently enforce inbound media byte limits before buffering remote media across multiple channel ingestion paths. This allows a remote sender to trigger oversized media downloads, causing elevated memory usage and potential process instability leading to denial of service. A fix is planned for version 2026.2.22.
A vulnerability in OpenClaw CLI versions up to 2026.2.24 allows reaction-only inbound Signal events to enqueue status events before sender access checks are applied. This flaw is limited to reaction notifications and does not affect normal direct message delivery or command execution. The issue was fixed in version 2026.2.25 by enforcing access checks before enqueuing reaction notifications.
An authorization mismatch in OpenClaw CLI allowed authenticated users with write-scoped agent access to invoke owner-only tools such as gateway and cron. This flaw permitted control-plane actions beyond the intended write scope. The issue affects versions up to 2026.2.26 and was fixed in version 2026.3.1 by enforcing consistent owner-only gating and tightening tool scope classification.
OpenClaw CLI versions up to 2026.2.23 contain a vulnerability in the system.run shell-wrapper where positional argv parameters can execute hidden commands that are not visible in the displayed approval text. This discrepancy allows execution of commands under misleading operator-visible text. The issue is fixed in version 2026.2.24.
OpenClaw CLI versions before 2026.2.23 have a vulnerability where Twilio webhook replay events could bypass the voice-call manager's deduplication mechanism. This happens because normalized event IDs were randomized on each parse, allowing replayed events to be treated as new and potentially causing duplicate or stale call-state transitions. The issue is fixed in version 2026.2.23 by preserving provider event IDs during normalization, adding bounded replay deduplication in webhook validation, and enforcing per-call turn-token checks.
OpenClaw CLI versions up to 2026.2.25 contain a workspace boundary bypass vulnerability. This flaw allows a symlink inside the workspace pointing to a non-existent target outside the workspace to bypass path validation on the first write, potentially creating files outside the intended workspace boundary. The issue is fixed in version 2026.2.26 by improving canonical path resolution to correctly evaluate missing leaf aliases. The vulnerability has a high severity with a CVSS score of 7.6.
OpenClaw CLI versions up to 2026.2.21-2 are vulnerable to an environment injection flaw that allows attacker-controlled shell startup files to execute before system.run's allowlist commands. This occurs because environment sanitization does not block certain startup override keys like HOME and ZDOTDIR, enabling remote code execution via crafted shell environments. A fix is available in version 2026.2.22 and later.
OpenClaw CLI versions up to 2026.2.24 have a trusted-proxy Control UI pairing bypass vulnerability (CVE-2026-32057). This flaw allows a websocket client with a node role to bypass pairing by setting client.id to 'control-ui', enabling unauthorized access to node event methods. The issue arises because the bypass did not require the operator role, weakening the authorization boundary. The vulnerability is patched in version 2026.2.25 by enforcing the operator role requirement for this bypass.
A medium-severity vulnerability in OpenClaw CLI affects approval-enabled workflows using host=node for system.run commands. The issue arises because system.run approvals did not enforce strict, versioned execution-context binding, allowing a previously approved request to be reused with altered environment input in certain configurations. The vulnerability requires specific conditions including exec approvals enabled, system.run usage via host=node, and access to an approval ID in the same context. A fix is planned in version 2026.2.26 that enforces versioned binding and removes legacy fallback matching.
OpenClaw CLI versions up to 2026.2.13 contain a path traversal vulnerability in the apply_patch feature that can write or delete files outside the intended workspace directory if filesystem sandbox containment is not enabled. This occurs because the path resolution does not enforce workspace containment when apply_patch is enabled and the agent runs without sandboxing. The vulnerability is fixed in version 2026.2.14. The impact depends on deployment specifics and who can trigger the tool execution. Mitigations include disabling apply_patch if not needed, keeping workspaceOnly containment enabled, and restricting tool execution permissions.