OpenClaw Multiple Authorization and Command Execution

An authorization mismatch vulnerability in OpenClaw versions prior to 2026.3.1 allows authenticated users with operator.write scope to escalate privileges and execute owner-only administrative functions including gateway and cron management through agent runs in scoped-token deployments. This is a privilege escalation issue affecting deployments using scoped authentication tokens, where write-level access can be exploited to perform control-plane operations reserved for owners. With a CVSS score of 8.8 and network-accessible attack vector, this represents a significant authorization bypass, though no KEV listing or public exploitation indicators are currently available.

Unauthenticated access to the noVNC observer interface in OpenClaw's Docker sandbox browser allows anyone who can reach the exposed port to view and interact with the sandboxed browser session without credentials. The flaw affects OpenClaw versions prior to 2026.2.21 (Docker image docker/openclaw <= 2026.2.19-2), where scripts/sandbox-browser-entrypoint.sh launched x11vnc with the -nopw flag, disabling VNC authentication entirely. Because the noVNC port is published to host loopback (127.0.0.1) by default, exploitation requires local presence on the host or an operator configuration that re-publishes the port to a broader interface; publicly available exploit code exists, though CISA KEV has not recorded active exploitation and EPSS is low at 0.04% (11th percentile).

OpenClaw versions before 2026.2.26 allow authenticated attackers to write arbitrary files outside the workspace directory by exploiting improper symlink resolution in path validation checks. An attacker with workspace access can leverage in-workspace symlinks pointing to external targets to bypass boundary restrictions on the first write operation. Public exploit code exists for this vulnerability, and a patch is available.

OpenClaw versions prior to 2026.2.22 contain a shell environment variable injection vulnerability in the system.run function that allows attackers to bypass command allowlist protections. Authenticated remote attackers with low privileges can inject malicious shell startup files (.bash_profile, .zshenv) via unsanitized HOME and ZDOTDIR variables to achieve arbitrary code execution before allowlisted commands execute. A patch is available from the vendor via GitHub commit c2c7114ed39a547ab6276e1e933029b9530ee906.

OpenClaw versions prior to 2026.2.22 contain a resource exhaustion vulnerability where the application fails to consistently enforce configured inbound media byte limits across multiple channel ingestion paths. Remote unauthenticated attackers can exploit this by sending oversized media payloads to cause elevated memory consumption and process instability, leading to denial of service. The vulnerability has a CVSS score of 7.5 (High severity) with network-based attack vector and low complexity, though no active exploitation (KEV) or public proof-of-concept has been reported at this time.

OpenClaw versions prior to 2026.3.1 contain a sandbox escape vulnerability that allows authenticated attackers with low privileges to bypass runtime confinement restrictions. Attackers can exploit a flaw in cross-agent sessions_spawn operations to create child processes under unsandboxed agents, effectively disabling sandbox protections by setting sandbox.mode to off. While the CVSS score is 7.5 (High), there is no evidence of active exploitation (not in CISA KEV), though the vulnerability has been publicly disclosed through GitHub Security Advisories and VulnCheck, increasing the likelihood of proof-of-concept development.

OpenClaw versions prior to 2026.2.23 contain a webhook event deduplication bypass vulnerability where normalized Twilio event IDs are randomized on each parse, allowing attackers to replay webhook events and circumvent the manager's deduplication checks. An unauthenticated remote attacker can exploit this over the network to trigger duplicate or stale call-state transitions, potentially causing incorrect call handling and state corruption. While no CVSS modifier for active exploitation or public POC is explicitly confirmed in the provided intelligence, the CVSS 6.5 score reflects moderate integrity and availability impact with low attack complexity.

OpenClaw versions prior to 2026.2.25 contain a symlink traversal vulnerability in the browser trace and download output path handling that allows local attackers with limited privileges to escape the managed temporary root directory and overwrite arbitrary files on the system. An attacker can create symbolic links to redirect file writes outside the intended sandbox, resulting in information disclosure and potential system compromise through arbitrary file modification. A patch is available from the vendor, and this vulnerability requires local access with low privileges to exploit, making it a medium-severity concern for multi-user systems.

OpenClaw versions before 2026.2.24 allow authenticated attackers to execute arbitrary commands through command injection in the system.run shell-wrapper by injecting malicious arguments that bypass validation controls. Public exploit code exists for this vulnerability, enabling attackers to disguise malicious payloads while executing hidden commands with the privileges of the affected application.

OpenClaw versions prior to 2026.2.21 contain a passwordless fallback authentication bypass in the BlueBubbles webhook handler that allows attackers to send unauthenticated webhook events by exploiting loopback or reverse-proxy heuristics. The vulnerability affects the BlueBubbles plugin component and has a CVSS score of 4.8 (medium severity) with low attack complexity, enabling both confidentiality and integrity impact without requiring authentication or user interaction. A vendor patch is available, and the vulnerability is documented in public advisories from VulnCheck and GitHub Security.

OpenClaw versions before 2026.2.25 allow authenticated attackers with node role permissions to bypass device pairing requirements in the Control UI by spoofing the control-ui client identifier, enabling unauthorized access to node event execution flows. Public exploit code exists for this authentication bypass vulnerability. The vulnerability requires prior authentication and has moderate integrity impact potential.

OpenClaw versions prior to 2026.2.26 contain an authentication bypass vulnerability in their Slack system event handlers that fails to properly enforce sender authorization checks. Attackers with low-privilege access (PR:L in CVSS vector) can craft and send unauthorized system events through message_changed, message_deleted, and thread_broadcast event types to bypass Slack DM allowlists and per-channel user allowlists. The vulnerability has a moderate CVSS score of 5.4 with low confidentiality and integrity impact; no KEV or active exploitation has been publicly disclosed, but a patch is available from the vendor.

OpenClaw versions prior to 2026.2.23 contain an authorization bypass vulnerability in the ACP (Approval Control Panel) client that automatically approves tool calls based on untrusted metadata and overly permissive heuristics. An authenticated attacker with PR (privileges required) can bypass interactive approval prompts for read-class operations by spoofing toolCall.kind metadata or using non-core read-like function names to reach auto-approve execution paths. This vulnerability enables unauthorized information disclosure and modification without user interaction, and while not currently listed as actively exploited in KEV, proof-of-concept demonstrations are available via vendor security advisories.

OpenClaw versions prior to 2026.2.21 contain an improper sandbox configuration vulnerability (CWE-1188) that allows local attackers with low privileges to execute arbitrary code on the host system by exploiting disabled OS-level sandbox protections in the Chromium browser container. The vulnerability does not require a sandbox escape, making exploitation straightforward for local users. A patch is available from the vendor, and the issue was reported by VulnCheck with references to GitHub security advisories and patch commits.

OpenClaw versions prior to 2026.2.25 contain an approval-integrity bypass vulnerability in the system.run function where the rendered command text displayed to approvers has whitespace trimmed from argv tokens, but the actual runtime execution uses the raw, untrimmed argv. An attacker with the ability to influence command arguments and reuse an approval context can craft a trailing-space executable token to execute a different binary than what was approved, resulting in arbitrary command execution under the OpenClaw runtime user. The CVSS score of 4.8 reflects the requirement for local privileges and user interaction, though the integrity impact is marked as high due to the ability to execute unauthorized commands.

OpenClaw fails to consistently apply sender-policy checks to reaction and pin event handlers, allowing authenticated attackers to bypass configured direct message policies and channel user allowlists by injecting unauthorized events from restricted senders. The vulnerability affects OpenClaw versions prior to 2026.2.25, requires low privileges (authenticated user), and enables unauthorized event injection with moderate severity (CVSS 4.3). A patch is available from the vendor, and the vulnerability has been documented in the VulnCheck advisory and GitHub Security Advisory GHSA-rm2p-j3r7-4x4j.

OpenClaw versions prior to 2026.2.25 contain an access control vulnerability in the signal reaction notification handling mechanism that allows unauthenticated attackers to enqueue status events before authorization checks are performed. Attackers can exploit the reaction-only event path in event-handler.ts to inject signal reaction status lines into sessions without validating proper DM or group access permissions, resulting in integrity compromise. The vulnerability has a CVSS score of 3.7 (low-to-moderate severity) with an attack vector of network, high complexity, and no privileges required, though no active exploitation or public proof-of-concept has been confirmed in known exploit databases.

OpenClaw versions prior to 2026.2.22 suffer from cryptographic secret reuse where the gateway authentication token is inappropriately reused as a fallback hashing secret for owner-ID obfuscation in system prompts sent to third-party model providers. An unauthenticated attacker with visibility into system prompts (such as through model provider logs, prompt injection, or interception) can reverse-engineer the gateway authentication token from hash outputs when commands.ownerDisplay is set to hash and commands.ownerDisplaySecret is unset, directly compromising authentication security. The vulnerability has a low CVSS score of 3.7 due to high attack complexity and limited impact scope, but represents a critical cryptographic design flaw that violates separation-of-concerns principles across security domains.

OpenClaw versions prior to 2026.2.26 contain an approval context-binding weakness that allows attackers to reuse previously approved system.run execution requests with modified environment variables, bypassing approval-enabled workflow integrity controls. An attacker with access to an approval ID can exploit this vulnerability to execute commands with different environment settings than originally approved, effectively circumventing execution-integrity safeguards. The vulnerability requires local/network access and user interaction, resulting in a low CVSS score of 2.6, but represents a meaningful integrity violation in approval workflows where execution consistency is critical.

OpenClaw versions prior to 2026.2.26 contain an authorization bypass vulnerability in the pairing-store access control mechanism for direct message pairing policies, allowing attackers to reuse pairing approvals across multiple accounts in multi-account deployments. An authenticated attacker (PR:L) who has been approved as a sender in one account can be automatically accepted in another account without explicit re-approval, effectively bypassing authorization boundaries between accounts. The vulnerability has a CVSS score of 3.7 with medium attack complexity and low confidentiality and integrity impacts; no active exploitation in the wild (KEV) or public proof-of-concept has been confirmed, but patches are available from the vendor.