Openclaw
Monthly
OpenClaw versions prior to 2026.2.25 contain an authorization bypass vulnerability in interactive callback handlers (block_action, view_submission, view_closed) that allows authenticated but unauthorized workspace members to bypass sender authorization checks and enqueue arbitrary system events into active sessions. This affects shared workspace deployments where multiple users with varying permission levels coexist, enabling privilege escalation and information disclosure attacks without requiring elevated privileges or user interaction.
OpenClaw prior to version 2026.3.2 allows unauthenticated attackers to bypass authentication controls on the /api/channels endpoint through path canonicalization mismatches, enabling access to protected API resources. The vulnerability exploits inconsistent handling of multi-encoded slash characters (%2f variants) between authentication checks and route processing. No patch is currently available, and exploitation requires only network access with no user interaction.
OpenClaw versions before 2026.2.22 allow high-privileged attackers to execute arbitrary shell commands by injecting malicious environment variables into the system.run function, bypassing the intended command allowlist protections. By exploiting bash xtrace expansion through SHELLOPTS and PS4 variables, an attacker with request-scoped environment variable access can achieve code execution beyond the restricted command set. No patch is currently available for this command injection vulnerability.
OpenClaw versions before 2026.2.23 allow authenticated users to bypass sandbox restrictions and read files outside the intended workspace by exploiting inadequate path validation in the sandboxed image tool. An attacker with valid credentials can exfiltrate sensitive files by leveraging vision model provider integrations, compromising the confidentiality of restricted data.
OpenClaw prior to version 2026.2.22 allows authenticated users to bypass device identity verification and assume a node role during WebSocket connections, enabling injection of unauthorized node events that trigger sensitive agent and voice transcript operations. An attacker with a shared gateway token can exploit this to perform actions without proper device pairing, potentially compromising system integrity and confidentiality. No patch is currently available.
Command injection in OpenClaw versions before 2026.2.19 allows local attackers with limited privileges to execute arbitrary commands when the Lobster extension tool falls back to Windows shell execution after subprocess failures. The vulnerability exists because the tool uses shell: true after spawn errors, enabling attackers to inject shell metacharacters into command arguments. A patch is available for affected users.
OpenClaw versions prior to 2026.3.1 contain a current working directory (cwd) injection vulnerability in Windows wrapper resolution for .cmd/.bat files that allows local attackers to manipulate command execution through directory control during shell fallback mechanisms. An authenticated local attacker with low privileges can exploit this vulnerability to achieve command execution integrity loss by controlling the working directory, potentially leading to unauthorized code execution or privilege escalation. While no active in-the-wild exploitation has been reported in KEV databases, the vulnerability is documented with a proof-of-concept available through the vendor's security advisory on GitHub.
OpenClaw versions prior to 2026.3.1 contain a post-approval executable rebind vulnerability in the system.run approval mechanism that fails to pin executable identity when argv[0] is not a full path. An attacker with local access and low privileges can modify PATH environment variables after an operator approves a command execution to redirect the approval to execute a different binary, achieving arbitrary command execution with the privileges of the OpenClaw process. The vulnerability has a moderate CVSS score of 6.0 reflecting local attack vector and high privilege requirements, but poses significant risk in environments where approval workflows are relied upon for security boundaries.
OpenClaw versions prior to 2026.2.19 contain an input validation bypass in the tools.exec.safeBins component that allows local attackers with command execution privileges to circumvent stdin-only restrictions and perform arbitrary filesystem operations. By exploiting sort output flags (specifically the -o flag for arbitrary file writes) or recursive grep flags (-R for recursive file reads), authenticated attackers can read sensitive files or overwrite critical files despite intended access controls. While the CVSS score of 3.6 is moderate and requires local access with low privileges, the vulnerability represents a privilege escalation or sandbox escape technique rather than a critical remote exploit.
OpenClaw versions prior to 2026.2.19 allow local attackers with limited privileges to execute arbitrary commands through the Lobster extension's Windows shell fallback mechanism by injecting malicious arguments into workflow processes. The vulnerability exploits cmd.exe command interpretation when spawn operations fail and trigger shell execution, enabling command injection with potential impact on system integrity and availability. A patch is available for affected versions.
OpenClaw versions before 2026.2.22 contain an allowlist parsing flaw in the macOS companion app that enables authenticated operators with elevated privileges to bypass command execution controls and run arbitrary commands on paired hosts. The vulnerability affects systems with operator.write access and macOS beta nodes, allowing attackers to craft malicious shell-chain payloads that circumvent validation checks. A security patch is available.
OpenClaw versions prior to 2026.2.26 contain an authorization bypass vulnerability in Signal group allowlist enforcement where the system incorrectly accepts sender identities derived from direct message (DM) pairing-store approvals. An authenticated attacker with low privileges can exploit this boundary weakness by obtaining DM pairing approval, allowing them to bypass group allowlist checks and gain unauthorized access to Signal groups. While the CVSS score is moderate (3.7) and attack complexity is high, the vulnerability represents a direct authentication control bypass in a messaging security context, and patches are available from the vendor.
OpenClaw versions prior to 2026.3.2 contain a symlink traversal vulnerability in the stageSandboxMedia function that fails to validate destination symlinks during media staging operations. This allows local attackers with low privileges to write files outside the intended sandbox workspace by placing malicious symlinks in the media/inbound directory, resulting in arbitrary file overwrite on the host system. A patch is available from the vendor, and the vulnerability was reported by VulnCheck with public references including a GitHub security advisory and commit fix.
OpenClaw 2026.3.1 contains an approval integrity bypass vulnerability in the system.run node-host execution feature where attackers can rewrite command-line arguments (argv) to change the semantics of operator-approved commands. An authenticated local attacker with low privileges can place malicious scripts in the working directory to execute unintended code despite the operator approving different command text, resulting in high-impact confidentiality, integrity, and availability violations. A patch is available from the vendor, and no public exploit code has been widely reported, but the vulnerability represents a critical trust boundary violation in approval workflows.
OpenClaw versions prior to 2026.2.22 contain an allowlist bypass vulnerability in the system.run function that allows authenticated attackers to execute non-allowlisted commands by exploiting shell line-continuation characters to fold malicious command substitution past security controls. An attacker with low privileges (PR:L) can inject shell metacharacters (specifically $\ followed by newline and parenthesis within double quotes) to circumvent approval boundaries and execute arbitrary commands, resulting in integrity compromise and potential availability impact. A public advisory and patch are available from the vendor, though no EPSS score or KEV status was provided in the intelligence sources.
OpenClaw versions prior to 2026.2.25 suffer from a webhook replay vulnerability where valid signed Nextcloud Talk webhook requests lack durable replay state suppression, allowing attackers to capture and replay previously legitimate signed requests to trigger duplicate inbound message processing. This can result in message duplication, data integrity issues, and potential availability degradation. While the CVSS score of 4.8 is moderate, the attack requires no authentication and can be executed over the network with medium complexity, making it a viable attack vector for threat actors with network visibility to webhook traffic.
OpenClaw versions before 2026.3.2 are vulnerable to a race condition in ZIP extraction that permits local attackers with limited privileges to write arbitrary files outside the intended extraction directory. By manipulating symlinks between path validation and write operations, an attacker can achieve arbitrary file placement on the system. A patch is available to resolve this integrity issue.
OpenClaw versions before 2026.2.19 allow local authenticated users to execute arbitrary commands by injecting shell metacharacters into environment variable values during Windows Scheduled Task script generation. The vulnerability stems from unquoted variable assignments in gateway.cmd that fail to sanitize special characters like &, |, ^, %, and !, enabling command injection when the task script runs. A patch is available to address this local privilege escalation risk.
OpenClaw versions prior to 2026.2.26 contain a Time-of-Check-Time-of-Use (TOCTOU) approval bypass vulnerability in the system.run execution function that allows local attackers with low privileges to execute arbitrary commands from unintended filesystem locations. An attacker can exploit a race condition by modifying parent symlinks in the current working directory after command approval but before execution, redirecting execution while maintaining the appearance of a safe working directory. A patch is available from the vendor, and this vulnerability has been documented by both VulnCheck and the OpenClaw security advisory (GHSA-f7ww-2725-qvw2).
OpenClaw versions prior to 2026.2.21 are vulnerable to prototype pollution attacks via the /debug set endpoint, allowing authenticated attackers to inject reserved prototype keys (__proto__, constructor, prototype) and manipulate object prototypes to bypass command gate restrictions. The vulnerability requires authenticated access and has relatively low real-world exploitability due to high attack complexity, but presents a meaningful integrity risk for authorized users who may not be aware of this attack vector. A patch is available from the vendor.
OpenClaw versions prior to 2026.2.24 contain a sandbox bind validation bypass vulnerability that allows local attackers with low privileges to circumvent allowed-root and blocked-path security checks through symlinked parent directories combined with non-existent leaf paths. An attacker can craft bind source paths that appear to reside within permitted sandbox roots but resolve outside sandbox boundaries once missing path components are created, effectively weakening the sandbox's bind-source isolation enforcement. A patch is available from the vendor, and exploitation requires local access with standard user privileges, making this a practical threat in multi-tenant or shared-system environments.
OpenClaw versions prior to 2026.2.24 contain a local media root bypass vulnerability that allows authenticated attackers to read arbitrary files from the host system through the sendAttachment and setGroupIcon message actions when sandboxRoot configuration is unset. An attacker with valid credentials can exploit path traversal to hydrate media from absolute file paths, gaining unauthorized access to sensitive files accessible by the OpenClaw runtime user. A patch is available from the vendor, and this vulnerability has been tracked in the ENISA EUVD database (EUVD-2026-12732) with confirmed GitHub security advisory and commit-level patch information.
OpenClaw versions prior to 2026.2.23 contain an arbitrary code execution vulnerability in shell-env that allows local attackers with low privileges to execute attacker-controlled binaries by manipulating the $SHELL environment variable through trusted-prefix fallback logic. An attacker who can write to directories like /opt/homebrew/bin can trick OpenClaw into executing malicious binaries in its process context, potentially escalating privileges or compromising system integrity. A patch is available from the vendor, and this vulnerability has been documented by VulnCheck and tracked under EUVD-2026-12730.
OpenClaw versions prior to 2026.3.2 contain a DNS pinning bypass vulnerability that allows authenticated attackers to circumvent SSRF (Server-Side Request Forgery) protections by exploiting environment proxy variable configuration. When HTTP_PROXY, HTTPS_PROXY, or ALL_PROXY environment variables are present, attackers can route malicious URLs through proxy mechanisms instead of pinned-destination routing, enabling access to internal resources that should be protected. The vulnerability requires low privilege (PR:L) and non-interactive attack (UI:N) with medium attack complexity (AC:H), resulting in high confidentiality impact (C:H) and lesser integrity and availability impact. A patch is available from the vendor.
OpenClaw prior to version 2026.3.2 allows local users with standard privileges to write files outside designated directories through insufficient path validation in the browser output handler. An attacker can exploit this path-confinement bypass to place malicious files in arbitrary filesystem locations, potentially leading to privilege escalation or system compromise.
OpenClaw versions before 2026.2.19 are vulnerable to regex injection and denial of service through unescaped Feishu mention metadata in the stripBotMention function. An unauthenticated network attacker can craft malicious mention metadata containing nested-quantifier patterns or regex metacharacters to trigger catastrophic backtracking, block message processing, or remove unintended content before model processing, with a CVSS score of 6.5 indicating medium severity with integrity and availability impact. Patch availability exists from the vendor via GitHub commits, and proof-of-concept details are available through VulnCheck advisory references.
OpenClaw versions prior to 2026.2.21 contain an environment variable injection vulnerability that allows authenticated local attackers to execute arbitrary code at startup time by injecting dangerous process-control variables (such as NODE_OPTIONS or LD_*) through the configuration env.vars mechanism. An attacker with local privileges can manipulate the gateway service's runtime environment to achieve code execution in the service context, potentially compromising the entire OpenClaw deployment. A patch is available from the vendor, and this vulnerability has been documented by VulnCheck with supporting references to the GitHub security advisory and corresponding commit fix.
OpenClaw Gateway versions prior to 2026.2.22 leak authentication tokens through Chrome DevTools Protocol (CDP) probe traffic on loopback interfaces, allowing local attackers to intercept the x-OpenClaw-relay-token header and reuse it for unauthorized Gateway access. An attacker with local network access or control of a loopback port can capture reachability probes to the /json/version endpoint and escalate privileges by replaying the stolen token as bearer authentication. A vendor patch is available, and this vulnerability has been documented by VulnCheck with references to the official GitHub security advisory and patch commit.
OpenClaw versions prior to 2026.2.22 contain an access control bypass vulnerability in the optional BlueBubbles plugin where empty allowFrom configuration causes the allowlist validation logic to fail, enabling remote attackers to send direct messages to BlueBubbles accounts without proper authorization. The vulnerability stems from improper handling of misconfigured sender authorization checks, allowing attackers to circumvent dmPolicy pairing and allowlist restrictions. Patches are available from the vendor, and this is classified as an authentication bypass issue with a CVSS score of 4.8 indicating moderate severity.
OpenClaw versions prior to 2026.2.21 contain an approval-integrity mismatch vulnerability in the system.run function that allows authenticated operators to execute arbitrary commands on Windows systems by appending malicious arguments after cmd.exe /c while the approval audit log records only the benign command text. An authenticated attacker with operator privileges can exploit this to achieve local command execution on trusted Windows nodes with mismatched or incomplete audit trails, enabling information disclosure and lateral movement while evading detection.
A vulnerability was identified in OpenClaw versions up to 2026.2.17. is affected by information exposure (CVSS 3.3).
Code injection in OpenClaw 2026.2.19 and earlier through the Skill Env Handler's applySkillConfigenvOverrides function allows authenticated remote attackers to execute arbitrary code with low integrity and confidentiality impact. An authenticated user can manipulate environment configuration settings to inject malicious code that executes in the context of the application. Mitigation requires upgrading to version 2026.2.21-beta.1 or later, as no official patch is currently available for production releases.
OpenClaw versions before 2026.2.17 allow privileged users with config modification access to read arbitrary files on the system through path traversal in the $include directive. An attacker in this position can exploit absolute paths, directory traversal sequences, or symlinks to access sensitive data like API keys and credentials that the OpenClaw process can read. No patch is currently available for this medium-severity vulnerability.
Openclaw versions up to 2026.2.12 is affected by missing authentication for critical function (CVSS 5.9).
Openclaw versions up to 2026.2.14 is affected by allocation of resources without limits or throttling (CVSS 5.5).
Openclaw versions up to 2026.2.14 is affected by missing authentication for critical function (CVSS 6.5).
OpenClaw versions 2026.1.16 through 2026.2.13 allow local attackers to write arbitrary files outside intended directories by supplying malicious archives to the skills, hooks, plugins, or signal installation commands. Successful exploitation enables attackers to achieve code execution or establish persistence on affected systems. A patch is available for affected users.
OpenClaw versions 2026.1.30 and earlier leak authentication bearer tokens to untrusted domains when the optional MS Teams attachment downloader extension is enabled, due to overly permissive suffix-based domain allowlisting during download retries. An attacker could harvest these tokens from allowed domains to compromise authenticated sessions. No patch is currently available, affecting users of the vulnerable versions.
OpenClaw prior to version 2026.2.14 fails to properly validate Telegram allowlist entries by accepting mutable usernames instead of immutable user IDs, enabling attackers to register recycled usernames and bypass access controls. An unauthenticated attacker can exploit this flaw to impersonate legitimate users and send unauthorized commands to OpenClaw bots. No patch is currently available.
OpenClaw versions before 2026.2.14 fail to validate base URLs in the Tlon Urbit extension, allowing attackers to trigger server-side request forgery attacks that direct the gateway to arbitrary hosts, including internal systems. This network-accessible vulnerability requires no authentication and can result in information disclosure and service disruption. No patch is currently available.
OpenClaw versions before 2026.2.13 are vulnerable to timing side-channel attacks on hook token validation due to use of non-constant-time string comparison. Remote attackers can exploit this weakness by measuring response times across multiple requests to gradually recover authentication tokens for the hooks endpoint. This affects confidentiality and integrity of OpenClaw deployments accessible over the network.
OpenClaw versions 2026.1.14-1 through 2026.2.1 with the Matrix plugin enabled fail to validate homeserver identity when checking direct message allowlists, allowing remote attackers to impersonate authorized users by spoofing display names or local identifiers from different homeservers. This bypass enables unauthorized access to routing and agent pipelines for authenticated Matrix users on remote servers. A patch is available in version 2026.2.2 and later.
OpenClaw prior to version 2026.2.2 is vulnerable to server-side request forgery through its attachment and media URL processing, allowing unauthenticated remote attackers to make arbitrary HTTP requests to internal resources. Attackers can exploit model-controlled message features to trigger the SSRF and exfiltrate response data as outbound attachments. Public exploit code exists for this vulnerability.
OpenClaw voice-call plugin versions before 2026.2.3 allow remote attackers to forge webhook events by exploiting improper authentication in reverse-proxy environments where forwarded headers are implicitly trusted. An unauthenticated attacker can manipulate Forwarded or X-Forwarded-* headers to bypass webhook verification and spoof legitimate events. A patch is available to address this authentication bypass vulnerability.
OpenClaw versions before 2026.2.12 are vulnerable to timing-based token extraction attacks due to non-constant-time string comparison in hook authentication. A network-based attacker can exploit this side-channel vulnerability to gradually recover the hook validation token through repeated timing measurements across multiple requests. The vulnerability requires repeated probing but poses a confidentiality risk to systems using vulnerable versions.
OpenClaw versions before 2026.2.14 allow local attackers to write arbitrary files outside the sandbox directory through path traversal sequences in crafted skill package names when sandbox skill mirroring is enabled. An attacker can exploit this by providing a malicious skill package with traversal patterns like ../ or absolute paths in the frontmatter name parameter, potentially compromising system integrity. A patch is available to remediate this vulnerability.
Openclaw versions up to 2026.2.14 is affected by allocation of resources without limits or throttling (CVSS 5.5).
OpenClaw versions before 2026.2.12 with the Nostr plugin enabled contain unauthenticated API endpoints that allow remote attackers to read and modify Nostr profiles without authentication when the gateway HTTP port is network-accessible. Attackers can exploit this to steal sensitive profile data, alter gateway configuration, and sign malicious Nostr events using the bot's private key. A patch is available for affected installations.
OpenClaw Chrome extension relay server versions prior to 2026.2.12 improperly bind to all network interfaces when wildcard cdpUrl values are configured, enabling remote attackers to discover service endpoints and port information. An attacker can exploit this exposure to conduct denial-of-service attacks and brute-force attempts against the relay token authentication mechanism without requiring local access.
Openclaw versions up to 2026.2.15 is affected by allocation of resources without limits or throttling (CVSS 6.5).
OpenClaw versions 2026.2.17 and earlier fail to enforce payload size limits in the ACP bridge, allowing local clients to trigger denial of service through excessively large prompt inputs that consume system resources. This vulnerability primarily impacts IDE integrations and other local ACP clients that may inadvertently send oversized text blocks. The issue has been patched in version 2026.2.19.
OpenClaw CLI versions 2026.2.13 and earlier terminate processes based on command-line pattern matching without verifying process ownership, allowing unrelated processes to be killed on shared hosts. An attacker or unprivileged user on a multi-tenant system could leverage this to disrupt services or cause denial of service by triggering process cleanup routines that match their target applications. The vulnerability has been patched in version 2026.2.14.
Openclaw contains a vulnerability that allows attackers to potential unintentional disclosure of local files from the packaging machine int (CVSS 4.4).
Unauthorized Discord moderation actions in OpenClaw versions 2026.2.17 and below allow non-admin users to execute timeouts, kicks, and bans by spoofing sender identity parameters in tool-driven requests. The vulnerability affects deployments where Discord moderation is enabled and the bot has necessary guild permissions, enabling privilege escalation through identity manipulation. A patch is available in version 2026.2.18.
OpenClaw versions prior to 2026.2.15 contain a stored XSS vulnerability in the Control UI where unsanitized assistant identity values (name/avatar) are injected into inline script tags, allowing authenticated attackers with high privileges to break out of the script context and execute arbitrary JavaScript. Public exploit code exists for this vulnerability. The issue has been remediated in version 2026.2.15 through removal of inline scripts and implementation of a restrictive Content Security Policy.
OpenClaw versions prior to 2026.2.15 allow authenticated administrators to write files outside the skill installation directory due to insufficient validation of the targetDir parameter during skill installation. An admin user could exploit this path traversal vulnerability to place malicious files in arbitrary locations on the system. A patch is available in version 2026.2.15 and later.
OpenClaw is a personal AI assistant. Prior to version 2026.2.15, `normalizeForHash` in `src/agents/sandbox/config-hash.ts` recursively sorted arrays that contained only primitive values. [CVSS 3.3 LOW]
OpenClaw AI assistant versions prior to 2026.2.15 allow local authenticated users to access session transcripts across peer accounts in multi-user shared-agent deployments due to insufficient session targeting restrictions. Additionally, Telegram webhook mode may fail to properly validate per-account secrets, potentially allowing unauthorized webhook access. The vulnerability primarily impacts multi-user environments with untrusted peers, while single-user or trusted deployments face limited practical risk.
OpenClaw versions prior to 2026.2.15 expose Telegram bot tokens in error messages and logs without redaction, allowing attackers who gain access to these logs to impersonate the bot and hijack its API access. This credential disclosure affects users of the AI assistant across systems where logs, crash reports, or support bundles are generated. Users must upgrade to version 2026.2.15 and rotate exposed Telegram bot tokens immediately.
OpenClaw versions 2026.1.12 through 2026.2.13 contain a path traversal vulnerability in the browser download helper that allows authenticated users with CLI access or valid gateway RPC tokens to write files outside the intended temporary downloads directory. An attacker with these credentials can exploit unsanitized output paths to place arbitrary files on the system. Version 2026.2.13 and later contain the fix.
OpenClaw versions prior to 2026.2.14 allow authenticated users to read arbitrary files from the Gateway host through path traversal in the browser tool's upload functionality. An attacker with valid Gateway credentials and browser tool permissions can supply absolute or traversal paths to bypass file access restrictions and access sensitive files. This vulnerability requires authentication and browser tool enablement but presents a high confidentiality risk to affected deployments.
OpenClaw versions prior to 2026.2.14 allow authenticated users to bypass group authorization policies by leveraging direct message trust credentials in group contexts, enabling unauthorized access to restricted group conversations. An attacker with valid credentials could exploit improper policy enforcement in iMessage groupPolicy=allowlist configurations to gain unauthorized visibility into protected group communications. A patch is available in version 2026.2.14 and later.
OpenClaw's mDNS/Bonjour discovery beacons transmit unauthenticated TXT records that iOS, macOS, and Android clients treat as authoritative for routing and TLS certificate pinning, allowing an attacker on a shared LAN to advertise a rogue service and redirect connections to attacker-controlled endpoints. An attacker can exploit this to bypass TLS pinning validation and potentially capture Gateway credentials through man-in-the-middle attacks. The vulnerability affects OpenClaw versions prior to 2026.2.14 and requires network proximity but no user interaction.
OpenClaw versions prior to 2026.2.14 expose sensitive configuration secrets through the skills.status endpoint to clients with operator.read privileges, allowing authenticated attackers to retrieve raw credential values including Discord tokens. The vulnerability affects AI/ML deployments where read-scoped access is intended to be non-sensitive; affected users should upgrade to version 2026.2.14 or later and rotate any exposed Discord tokens.
OpenClaw macOS desktop client versions 2026.2.6 through 2026.2.13 fail to fully display message content in confirmation dialogs for deep links, allowing attackers to hide malicious payloads behind whitespace that users cannot see before execution. When a user approves the truncated preview and clicks "Run," the full hidden message executes, potentially leading to arbitrary command execution depending on the user's configured permissions. This affects beta versions of the OpenClaw AI assistant on macOS where the openclaw:// URL scheme is registered without proper authentication.
OpenClaw (formerly Clawdbot) is a personal AI assistant users run on their own devices. In versions 2026.2.2 and below, when the Slack integration is enabled, channel metadata (topic/description) can be incorporated into the model's system prompt. [CVSS 3.7 LOW]
OpenClaw versions prior to 2026.1.30 suffer from a path traversal vulnerability in the isValidMedia() function that permits authenticated agents to read arbitrary files on the system by crafting malicious MEDIA output directives. An attacker with agent access can leverage this flaw to exfiltrate sensitive data accessible to the application process. Public exploit code exists for this vulnerability, and no patch is currently available.
OpenClaw versions prior to 2026.2.25 contain an authorization bypass vulnerability in interactive callback handlers (block_action, view_submission, view_closed) that allows authenticated but unauthorized workspace members to bypass sender authorization checks and enqueue arbitrary system events into active sessions. This affects shared workspace deployments where multiple users with varying permission levels coexist, enabling privilege escalation and information disclosure attacks without requiring elevated privileges or user interaction.
OpenClaw prior to version 2026.3.2 allows unauthenticated attackers to bypass authentication controls on the /api/channels endpoint through path canonicalization mismatches, enabling access to protected API resources. The vulnerability exploits inconsistent handling of multi-encoded slash characters (%2f variants) between authentication checks and route processing. No patch is currently available, and exploitation requires only network access with no user interaction.
OpenClaw versions before 2026.2.22 allow high-privileged attackers to execute arbitrary shell commands by injecting malicious environment variables into the system.run function, bypassing the intended command allowlist protections. By exploiting bash xtrace expansion through SHELLOPTS and PS4 variables, an attacker with request-scoped environment variable access can achieve code execution beyond the restricted command set. No patch is currently available for this command injection vulnerability.
OpenClaw versions before 2026.2.23 allow authenticated users to bypass sandbox restrictions and read files outside the intended workspace by exploiting inadequate path validation in the sandboxed image tool. An attacker with valid credentials can exfiltrate sensitive files by leveraging vision model provider integrations, compromising the confidentiality of restricted data.
OpenClaw prior to version 2026.2.22 allows authenticated users to bypass device identity verification and assume a node role during WebSocket connections, enabling injection of unauthorized node events that trigger sensitive agent and voice transcript operations. An attacker with a shared gateway token can exploit this to perform actions without proper device pairing, potentially compromising system integrity and confidentiality. No patch is currently available.
Command injection in OpenClaw versions before 2026.2.19 allows local attackers with limited privileges to execute arbitrary commands when the Lobster extension tool falls back to Windows shell execution after subprocess failures. The vulnerability exists because the tool uses shell: true after spawn errors, enabling attackers to inject shell metacharacters into command arguments. A patch is available for affected users.
OpenClaw versions prior to 2026.3.1 contain a current working directory (cwd) injection vulnerability in Windows wrapper resolution for .cmd/.bat files that allows local attackers to manipulate command execution through directory control during shell fallback mechanisms. An authenticated local attacker with low privileges can exploit this vulnerability to achieve command execution integrity loss by controlling the working directory, potentially leading to unauthorized code execution or privilege escalation. While no active in-the-wild exploitation has been reported in KEV databases, the vulnerability is documented with a proof-of-concept available through the vendor's security advisory on GitHub.
OpenClaw versions prior to 2026.3.1 contain a post-approval executable rebind vulnerability in the system.run approval mechanism that fails to pin executable identity when argv[0] is not a full path. An attacker with local access and low privileges can modify PATH environment variables after an operator approves a command execution to redirect the approval to execute a different binary, achieving arbitrary command execution with the privileges of the OpenClaw process. The vulnerability has a moderate CVSS score of 6.0 reflecting local attack vector and high privilege requirements, but poses significant risk in environments where approval workflows are relied upon for security boundaries.
OpenClaw versions prior to 2026.2.19 contain an input validation bypass in the tools.exec.safeBins component that allows local attackers with command execution privileges to circumvent stdin-only restrictions and perform arbitrary filesystem operations. By exploiting sort output flags (specifically the -o flag for arbitrary file writes) or recursive grep flags (-R for recursive file reads), authenticated attackers can read sensitive files or overwrite critical files despite intended access controls. While the CVSS score of 3.6 is moderate and requires local access with low privileges, the vulnerability represents a privilege escalation or sandbox escape technique rather than a critical remote exploit.
OpenClaw versions prior to 2026.2.19 allow local attackers with limited privileges to execute arbitrary commands through the Lobster extension's Windows shell fallback mechanism by injecting malicious arguments into workflow processes. The vulnerability exploits cmd.exe command interpretation when spawn operations fail and trigger shell execution, enabling command injection with potential impact on system integrity and availability. A patch is available for affected versions.
OpenClaw versions before 2026.2.22 contain an allowlist parsing flaw in the macOS companion app that enables authenticated operators with elevated privileges to bypass command execution controls and run arbitrary commands on paired hosts. The vulnerability affects systems with operator.write access and macOS beta nodes, allowing attackers to craft malicious shell-chain payloads that circumvent validation checks. A security patch is available.
OpenClaw versions prior to 2026.2.26 contain an authorization bypass vulnerability in Signal group allowlist enforcement where the system incorrectly accepts sender identities derived from direct message (DM) pairing-store approvals. An authenticated attacker with low privileges can exploit this boundary weakness by obtaining DM pairing approval, allowing them to bypass group allowlist checks and gain unauthorized access to Signal groups. While the CVSS score is moderate (3.7) and attack complexity is high, the vulnerability represents a direct authentication control bypass in a messaging security context, and patches are available from the vendor.
OpenClaw versions prior to 2026.3.2 contain a symlink traversal vulnerability in the stageSandboxMedia function that fails to validate destination symlinks during media staging operations. This allows local attackers with low privileges to write files outside the intended sandbox workspace by placing malicious symlinks in the media/inbound directory, resulting in arbitrary file overwrite on the host system. A patch is available from the vendor, and the vulnerability was reported by VulnCheck with public references including a GitHub security advisory and commit fix.
OpenClaw 2026.3.1 contains an approval integrity bypass vulnerability in the system.run node-host execution feature where attackers can rewrite command-line arguments (argv) to change the semantics of operator-approved commands. An authenticated local attacker with low privileges can place malicious scripts in the working directory to execute unintended code despite the operator approving different command text, resulting in high-impact confidentiality, integrity, and availability violations. A patch is available from the vendor, and no public exploit code has been widely reported, but the vulnerability represents a critical trust boundary violation in approval workflows.
OpenClaw versions prior to 2026.2.22 contain an allowlist bypass vulnerability in the system.run function that allows authenticated attackers to execute non-allowlisted commands by exploiting shell line-continuation characters to fold malicious command substitution past security controls. An attacker with low privileges (PR:L) can inject shell metacharacters (specifically $\ followed by newline and parenthesis within double quotes) to circumvent approval boundaries and execute arbitrary commands, resulting in integrity compromise and potential availability impact. A public advisory and patch are available from the vendor, though no EPSS score or KEV status was provided in the intelligence sources.
OpenClaw versions prior to 2026.2.25 suffer from a webhook replay vulnerability where valid signed Nextcloud Talk webhook requests lack durable replay state suppression, allowing attackers to capture and replay previously legitimate signed requests to trigger duplicate inbound message processing. This can result in message duplication, data integrity issues, and potential availability degradation. While the CVSS score of 4.8 is moderate, the attack requires no authentication and can be executed over the network with medium complexity, making it a viable attack vector for threat actors with network visibility to webhook traffic.
OpenClaw versions before 2026.3.2 are vulnerable to a race condition in ZIP extraction that permits local attackers with limited privileges to write arbitrary files outside the intended extraction directory. By manipulating symlinks between path validation and write operations, an attacker can achieve arbitrary file placement on the system. A patch is available to resolve this integrity issue.
OpenClaw versions before 2026.2.19 allow local authenticated users to execute arbitrary commands by injecting shell metacharacters into environment variable values during Windows Scheduled Task script generation. The vulnerability stems from unquoted variable assignments in gateway.cmd that fail to sanitize special characters like &, |, ^, %, and !, enabling command injection when the task script runs. A patch is available to address this local privilege escalation risk.
OpenClaw versions prior to 2026.2.26 contain a Time-of-Check-Time-of-Use (TOCTOU) approval bypass vulnerability in the system.run execution function that allows local attackers with low privileges to execute arbitrary commands from unintended filesystem locations. An attacker can exploit a race condition by modifying parent symlinks in the current working directory after command approval but before execution, redirecting execution while maintaining the appearance of a safe working directory. A patch is available from the vendor, and this vulnerability has been documented by both VulnCheck and the OpenClaw security advisory (GHSA-f7ww-2725-qvw2).
OpenClaw versions prior to 2026.2.21 are vulnerable to prototype pollution attacks via the /debug set endpoint, allowing authenticated attackers to inject reserved prototype keys (__proto__, constructor, prototype) and manipulate object prototypes to bypass command gate restrictions. The vulnerability requires authenticated access and has relatively low real-world exploitability due to high attack complexity, but presents a meaningful integrity risk for authorized users who may not be aware of this attack vector. A patch is available from the vendor.
OpenClaw versions prior to 2026.2.24 contain a sandbox bind validation bypass vulnerability that allows local attackers with low privileges to circumvent allowed-root and blocked-path security checks through symlinked parent directories combined with non-existent leaf paths. An attacker can craft bind source paths that appear to reside within permitted sandbox roots but resolve outside sandbox boundaries once missing path components are created, effectively weakening the sandbox's bind-source isolation enforcement. A patch is available from the vendor, and exploitation requires local access with standard user privileges, making this a practical threat in multi-tenant or shared-system environments.
OpenClaw versions prior to 2026.2.24 contain a local media root bypass vulnerability that allows authenticated attackers to read arbitrary files from the host system through the sendAttachment and setGroupIcon message actions when sandboxRoot configuration is unset. An attacker with valid credentials can exploit path traversal to hydrate media from absolute file paths, gaining unauthorized access to sensitive files accessible by the OpenClaw runtime user. A patch is available from the vendor, and this vulnerability has been tracked in the ENISA EUVD database (EUVD-2026-12732) with confirmed GitHub security advisory and commit-level patch information.
OpenClaw versions prior to 2026.2.23 contain an arbitrary code execution vulnerability in shell-env that allows local attackers with low privileges to execute attacker-controlled binaries by manipulating the $SHELL environment variable through trusted-prefix fallback logic. An attacker who can write to directories like /opt/homebrew/bin can trick OpenClaw into executing malicious binaries in its process context, potentially escalating privileges or compromising system integrity. A patch is available from the vendor, and this vulnerability has been documented by VulnCheck and tracked under EUVD-2026-12730.
OpenClaw versions prior to 2026.3.2 contain a DNS pinning bypass vulnerability that allows authenticated attackers to circumvent SSRF (Server-Side Request Forgery) protections by exploiting environment proxy variable configuration. When HTTP_PROXY, HTTPS_PROXY, or ALL_PROXY environment variables are present, attackers can route malicious URLs through proxy mechanisms instead of pinned-destination routing, enabling access to internal resources that should be protected. The vulnerability requires low privilege (PR:L) and non-interactive attack (UI:N) with medium attack complexity (AC:H), resulting in high confidentiality impact (C:H) and lesser integrity and availability impact. A patch is available from the vendor.
OpenClaw prior to version 2026.3.2 allows local users with standard privileges to write files outside designated directories through insufficient path validation in the browser output handler. An attacker can exploit this path-confinement bypass to place malicious files in arbitrary filesystem locations, potentially leading to privilege escalation or system compromise.
OpenClaw versions before 2026.2.19 are vulnerable to regex injection and denial of service through unescaped Feishu mention metadata in the stripBotMention function. An unauthenticated network attacker can craft malicious mention metadata containing nested-quantifier patterns or regex metacharacters to trigger catastrophic backtracking, block message processing, or remove unintended content before model processing, with a CVSS score of 6.5 indicating medium severity with integrity and availability impact. Patch availability exists from the vendor via GitHub commits, and proof-of-concept details are available through VulnCheck advisory references.
OpenClaw versions prior to 2026.2.21 contain an environment variable injection vulnerability that allows authenticated local attackers to execute arbitrary code at startup time by injecting dangerous process-control variables (such as NODE_OPTIONS or LD_*) through the configuration env.vars mechanism. An attacker with local privileges can manipulate the gateway service's runtime environment to achieve code execution in the service context, potentially compromising the entire OpenClaw deployment. A patch is available from the vendor, and this vulnerability has been documented by VulnCheck with supporting references to the GitHub security advisory and corresponding commit fix.
OpenClaw Gateway versions prior to 2026.2.22 leak authentication tokens through Chrome DevTools Protocol (CDP) probe traffic on loopback interfaces, allowing local attackers to intercept the x-OpenClaw-relay-token header and reuse it for unauthorized Gateway access. An attacker with local network access or control of a loopback port can capture reachability probes to the /json/version endpoint and escalate privileges by replaying the stolen token as bearer authentication. A vendor patch is available, and this vulnerability has been documented by VulnCheck with references to the official GitHub security advisory and patch commit.
OpenClaw versions prior to 2026.2.22 contain an access control bypass vulnerability in the optional BlueBubbles plugin where empty allowFrom configuration causes the allowlist validation logic to fail, enabling remote attackers to send direct messages to BlueBubbles accounts without proper authorization. The vulnerability stems from improper handling of misconfigured sender authorization checks, allowing attackers to circumvent dmPolicy pairing and allowlist restrictions. Patches are available from the vendor, and this is classified as an authentication bypass issue with a CVSS score of 4.8 indicating moderate severity.
OpenClaw versions prior to 2026.2.21 contain an approval-integrity mismatch vulnerability in the system.run function that allows authenticated operators to execute arbitrary commands on Windows systems by appending malicious arguments after cmd.exe /c while the approval audit log records only the benign command text. An authenticated attacker with operator privileges can exploit this to achieve local command execution on trusted Windows nodes with mismatched or incomplete audit trails, enabling information disclosure and lateral movement while evading detection.
A vulnerability was identified in OpenClaw versions up to 2026.2.17. is affected by information exposure (CVSS 3.3).
Code injection in OpenClaw 2026.2.19 and earlier through the Skill Env Handler's applySkillConfigenvOverrides function allows authenticated remote attackers to execute arbitrary code with low integrity and confidentiality impact. An authenticated user can manipulate environment configuration settings to inject malicious code that executes in the context of the application. Mitigation requires upgrading to version 2026.2.21-beta.1 or later, as no official patch is currently available for production releases.
OpenClaw versions before 2026.2.17 allow privileged users with config modification access to read arbitrary files on the system through path traversal in the $include directive. An attacker in this position can exploit absolute paths, directory traversal sequences, or symlinks to access sensitive data like API keys and credentials that the OpenClaw process can read. No patch is currently available for this medium-severity vulnerability.
Openclaw versions up to 2026.2.12 is affected by missing authentication for critical function (CVSS 5.9).
Openclaw versions up to 2026.2.14 is affected by allocation of resources without limits or throttling (CVSS 5.5).
Openclaw versions up to 2026.2.14 is affected by missing authentication for critical function (CVSS 6.5).
OpenClaw versions 2026.1.16 through 2026.2.13 allow local attackers to write arbitrary files outside intended directories by supplying malicious archives to the skills, hooks, plugins, or signal installation commands. Successful exploitation enables attackers to achieve code execution or establish persistence on affected systems. A patch is available for affected users.
OpenClaw versions 2026.1.30 and earlier leak authentication bearer tokens to untrusted domains when the optional MS Teams attachment downloader extension is enabled, due to overly permissive suffix-based domain allowlisting during download retries. An attacker could harvest these tokens from allowed domains to compromise authenticated sessions. No patch is currently available, affecting users of the vulnerable versions.
OpenClaw prior to version 2026.2.14 fails to properly validate Telegram allowlist entries by accepting mutable usernames instead of immutable user IDs, enabling attackers to register recycled usernames and bypass access controls. An unauthenticated attacker can exploit this flaw to impersonate legitimate users and send unauthorized commands to OpenClaw bots. No patch is currently available.
OpenClaw versions before 2026.2.14 fail to validate base URLs in the Tlon Urbit extension, allowing attackers to trigger server-side request forgery attacks that direct the gateway to arbitrary hosts, including internal systems. This network-accessible vulnerability requires no authentication and can result in information disclosure and service disruption. No patch is currently available.
OpenClaw versions before 2026.2.13 are vulnerable to timing side-channel attacks on hook token validation due to use of non-constant-time string comparison. Remote attackers can exploit this weakness by measuring response times across multiple requests to gradually recover authentication tokens for the hooks endpoint. This affects confidentiality and integrity of OpenClaw deployments accessible over the network.
OpenClaw versions 2026.1.14-1 through 2026.2.1 with the Matrix plugin enabled fail to validate homeserver identity when checking direct message allowlists, allowing remote attackers to impersonate authorized users by spoofing display names or local identifiers from different homeservers. This bypass enables unauthorized access to routing and agent pipelines for authenticated Matrix users on remote servers. A patch is available in version 2026.2.2 and later.
OpenClaw prior to version 2026.2.2 is vulnerable to server-side request forgery through its attachment and media URL processing, allowing unauthenticated remote attackers to make arbitrary HTTP requests to internal resources. Attackers can exploit model-controlled message features to trigger the SSRF and exfiltrate response data as outbound attachments. Public exploit code exists for this vulnerability.
OpenClaw voice-call plugin versions before 2026.2.3 allow remote attackers to forge webhook events by exploiting improper authentication in reverse-proxy environments where forwarded headers are implicitly trusted. An unauthenticated attacker can manipulate Forwarded or X-Forwarded-* headers to bypass webhook verification and spoof legitimate events. A patch is available to address this authentication bypass vulnerability.
OpenClaw versions before 2026.2.12 are vulnerable to timing-based token extraction attacks due to non-constant-time string comparison in hook authentication. A network-based attacker can exploit this side-channel vulnerability to gradually recover the hook validation token through repeated timing measurements across multiple requests. The vulnerability requires repeated probing but poses a confidentiality risk to systems using vulnerable versions.
OpenClaw versions before 2026.2.14 allow local attackers to write arbitrary files outside the sandbox directory through path traversal sequences in crafted skill package names when sandbox skill mirroring is enabled. An attacker can exploit this by providing a malicious skill package with traversal patterns like ../ or absolute paths in the frontmatter name parameter, potentially compromising system integrity. A patch is available to remediate this vulnerability.
Openclaw versions up to 2026.2.14 is affected by allocation of resources without limits or throttling (CVSS 5.5).
OpenClaw versions before 2026.2.12 with the Nostr plugin enabled contain unauthenticated API endpoints that allow remote attackers to read and modify Nostr profiles without authentication when the gateway HTTP port is network-accessible. Attackers can exploit this to steal sensitive profile data, alter gateway configuration, and sign malicious Nostr events using the bot's private key. A patch is available for affected installations.
OpenClaw Chrome extension relay server versions prior to 2026.2.12 improperly bind to all network interfaces when wildcard cdpUrl values are configured, enabling remote attackers to discover service endpoints and port information. An attacker can exploit this exposure to conduct denial-of-service attacks and brute-force attempts against the relay token authentication mechanism without requiring local access.
Openclaw versions up to 2026.2.15 is affected by allocation of resources without limits or throttling (CVSS 6.5).
OpenClaw versions 2026.2.17 and earlier fail to enforce payload size limits in the ACP bridge, allowing local clients to trigger denial of service through excessively large prompt inputs that consume system resources. This vulnerability primarily impacts IDE integrations and other local ACP clients that may inadvertently send oversized text blocks. The issue has been patched in version 2026.2.19.
OpenClaw CLI versions 2026.2.13 and earlier terminate processes based on command-line pattern matching without verifying process ownership, allowing unrelated processes to be killed on shared hosts. An attacker or unprivileged user on a multi-tenant system could leverage this to disrupt services or cause denial of service by triggering process cleanup routines that match their target applications. The vulnerability has been patched in version 2026.2.14.
Openclaw contains a vulnerability that allows attackers to potential unintentional disclosure of local files from the packaging machine int (CVSS 4.4).
Unauthorized Discord moderation actions in OpenClaw versions 2026.2.17 and below allow non-admin users to execute timeouts, kicks, and bans by spoofing sender identity parameters in tool-driven requests. The vulnerability affects deployments where Discord moderation is enabled and the bot has necessary guild permissions, enabling privilege escalation through identity manipulation. A patch is available in version 2026.2.18.
OpenClaw versions prior to 2026.2.15 contain a stored XSS vulnerability in the Control UI where unsanitized assistant identity values (name/avatar) are injected into inline script tags, allowing authenticated attackers with high privileges to break out of the script context and execute arbitrary JavaScript. Public exploit code exists for this vulnerability. The issue has been remediated in version 2026.2.15 through removal of inline scripts and implementation of a restrictive Content Security Policy.
OpenClaw versions prior to 2026.2.15 allow authenticated administrators to write files outside the skill installation directory due to insufficient validation of the targetDir parameter during skill installation. An admin user could exploit this path traversal vulnerability to place malicious files in arbitrary locations on the system. A patch is available in version 2026.2.15 and later.
OpenClaw is a personal AI assistant. Prior to version 2026.2.15, `normalizeForHash` in `src/agents/sandbox/config-hash.ts` recursively sorted arrays that contained only primitive values. [CVSS 3.3 LOW]
OpenClaw AI assistant versions prior to 2026.2.15 allow local authenticated users to access session transcripts across peer accounts in multi-user shared-agent deployments due to insufficient session targeting restrictions. Additionally, Telegram webhook mode may fail to properly validate per-account secrets, potentially allowing unauthorized webhook access. The vulnerability primarily impacts multi-user environments with untrusted peers, while single-user or trusted deployments face limited practical risk.
OpenClaw versions prior to 2026.2.15 expose Telegram bot tokens in error messages and logs without redaction, allowing attackers who gain access to these logs to impersonate the bot and hijack its API access. This credential disclosure affects users of the AI assistant across systems where logs, crash reports, or support bundles are generated. Users must upgrade to version 2026.2.15 and rotate exposed Telegram bot tokens immediately.
OpenClaw versions 2026.1.12 through 2026.2.13 contain a path traversal vulnerability in the browser download helper that allows authenticated users with CLI access or valid gateway RPC tokens to write files outside the intended temporary downloads directory. An attacker with these credentials can exploit unsanitized output paths to place arbitrary files on the system. Version 2026.2.13 and later contain the fix.
OpenClaw versions prior to 2026.2.14 allow authenticated users to read arbitrary files from the Gateway host through path traversal in the browser tool's upload functionality. An attacker with valid Gateway credentials and browser tool permissions can supply absolute or traversal paths to bypass file access restrictions and access sensitive files. This vulnerability requires authentication and browser tool enablement but presents a high confidentiality risk to affected deployments.
OpenClaw versions prior to 2026.2.14 allow authenticated users to bypass group authorization policies by leveraging direct message trust credentials in group contexts, enabling unauthorized access to restricted group conversations. An attacker with valid credentials could exploit improper policy enforcement in iMessage groupPolicy=allowlist configurations to gain unauthorized visibility into protected group communications. A patch is available in version 2026.2.14 and later.
OpenClaw's mDNS/Bonjour discovery beacons transmit unauthenticated TXT records that iOS, macOS, and Android clients treat as authoritative for routing and TLS certificate pinning, allowing an attacker on a shared LAN to advertise a rogue service and redirect connections to attacker-controlled endpoints. An attacker can exploit this to bypass TLS pinning validation and potentially capture Gateway credentials through man-in-the-middle attacks. The vulnerability affects OpenClaw versions prior to 2026.2.14 and requires network proximity but no user interaction.
OpenClaw versions prior to 2026.2.14 expose sensitive configuration secrets through the skills.status endpoint to clients with operator.read privileges, allowing authenticated attackers to retrieve raw credential values including Discord tokens. The vulnerability affects AI/ML deployments where read-scoped access is intended to be non-sensitive; affected users should upgrade to version 2026.2.14 or later and rotate any exposed Discord tokens.
OpenClaw macOS desktop client versions 2026.2.6 through 2026.2.13 fail to fully display message content in confirmation dialogs for deep links, allowing attackers to hide malicious payloads behind whitespace that users cannot see before execution. When a user approves the truncated preview and clicks "Run," the full hidden message executes, potentially leading to arbitrary command execution depending on the user's configured permissions. This affects beta versions of the OpenClaw AI assistant on macOS where the openclaw:// URL scheme is registered without proper authentication.
OpenClaw (formerly Clawdbot) is a personal AI assistant users run on their own devices. In versions 2026.2.2 and below, when the Slack integration is enabled, channel metadata (topic/description) can be incorporated into the model's system prompt. [CVSS 3.7 LOW]
OpenClaw versions prior to 2026.1.30 suffer from a path traversal vulnerability in the isValidMedia() function that permits authenticated agents to read arbitrary files on the system by crafting malicious MEDIA output directives. An attacker with agent access can leverage this flaw to exfiltrate sensitive data accessible to the application process. Public exploit code exists for this vulnerability, and no patch is currently available.