OpenClaw CLI versions from 2026.4.7 up to but not including 2026.4.20 are vulnerable to an environment variable injection issue where a malicious workspace .env file can override the MINIMAX_API_HOST. This can redirect credentialed MiniMax API requests to an attacker-controlled server, exposing the API key in the Authorization header. The vulnerability requires running OpenClaw from an attacker-controlled workspace. The issue is fixed in version 2026.4.20.
A vulnerability in OpenClaw CLI versions before 2026.4.20 causes Feishu card-action callbacks to misclassify direct messages (DMs) as group conversations. This misclassification leads to skipping enforcement of restrictive DM policies, allowing senders in Feishu DMs to trigger card-action flows that should be blocked. The issue is specific to Feishu card-action handling and has been fixed in version 2026.4.20.
OpenClaw CLI versions up to 2026.4.21 have a vulnerability where the Gateway Control UI bootstrap config endpoint does not require Gateway authentication, potentially exposing sensitive configuration fields. This issue is fixed in version 2026.4.22 by enforcing authentication on the bootstrap config route, aligning it with other authenticated Control UI reads.
OpenClaw CLI versions before 2026.4.20 contain a vulnerability where the MCP stdio server environment can load dangerous startup variables from workspace configuration. This allows malicious workspaces to pass harmful environment variables like NODE_OPTIONS, LD_PRELOAD, or BASH_ENV to the MCP server process, potentially causing it to load attacker-controlled code. The risk is limited to local or workspace trust boundaries and requires the operator to run OpenClaw in a maliciously configured workspace. The issue is fixed in version 2026.4.20 by filtering unsafe environment variables before spawning MCP servers.
OpenClaw CLI versions before 2026.4.15 have a vulnerability where the webchat audio embedding feature could read local files without enforcing local-root containment. This allows an attacker who can influence media URLs in webchat replies to cause the system to read and embed local audio files into responses. The vulnerability requires the file to be readable by the gateway process, have an audio-like extension, and fit within a size limit. The issue was fixed in version 2026.4.15 and later.
OpenClaw CLI versions up to 2026.4.21 have a vulnerability where ACP child sessions do not properly inherit subagent security envelope constraints. This can cause restrictions such as depth limits, child-count limits, control scope, or target-agent restrictions to be bypassed in child sessions. The issue is fixed in version 2026.4.22, which enforces these constraints correctly for child ACP sessions.
OpenClaw CLI versions prior to 2026.4.20 contain a local agent policy enforcement bypass vulnerability. Bundled MCP and LSP tools could bypass restrictive tool policies configured by an operator, such as allow/deny lists or sandbox restrictions. This could allow tools that should have been denied by policy to remain available. The issue requires a configured bundled MCP or LSP tool source and restrictive local policy. It does not allow unauthenticated remote compromise. The vulnerability is fixed in version 2026.4.20.
OpenClaw CLI versions prior to 2026.4.20 improperly labeled isolated cron awareness events as trusted system events instead of untrusted. This trust-labeling flaw could increase the impact of prompt injections but does not bypass authentication, tool policies, or sandboxing. The issue is fixed in version 2026.4.20 by preserving untrusted labels for these events.
OpenClaw CLI versions prior to 2026.4.20 have a vulnerability where browser CDP profile creation skips strict-mode SSRF checks. This allows stored profiles to reference private-network or metadata endpoints even when strict SSRF policies are enabled, potentially exposing internal resources during profile status checks. The issue only affects deployments with strict SSRF mode enabled, as default trusted-operator behavior permits private-network CDP endpoints. The vulnerability is considered low severity and has been fixed in version 2026.4.20.
OpenClaw CLI versions prior to 2026.4.20 contain a vulnerability where the agent gateway configuration mutation process could bypass protections on operator-trusted settings. This allows a model with access to the owner-only gateway tool to persist unauthorized changes to sensitive settings such as sandbox policy, plugin enablement, gateway authentication/TLS, hook routing, MCP server configuration, SSRF policy, and filesystem hardening. The vulnerability is a model-to-operator guard bypass and does not allow remote unauthenticated compromise. The issue is fixed in version 2026.4.20.
OpenClaw CLI versions up to 2026.4.21 allow workspace dotenv files to override connector endpoint hosts for services like Matrix, Mattermost, IRC, or Synology. This could redirect runtime traffic away from the operator-configured endpoints. The issue is fixed in version 2026.4.22 by blocking these endpoint variables in workspace dotenv files while keeping trusted global dotenv loading separate.
OpenClaw CLI versions before 2026.4.23 contain a vulnerability where the bundled plugin setup resolver could execute arbitrary JavaScript from an attacker-controlled setup-api.js file located in the current working directory. This occurs when a user runs OpenClaw commands from a directory containing malicious setup files, leading to code execution under the user's privileges. The vulnerability requires user interaction and local access but does not depend on network exposure. The issue is fixed in version 2026.4.23 by restricting setup resolution to the canonical package root and excluding process.cwd().
OpenClaw CLI versions before 2026.4.23 have a vulnerability where webhook route secrets backed by SecretRef remain cached after rotation and reload, allowing old secrets to remain valid until a plugin or gateway restart. This weakens credential rotation by permitting continued use of stale secrets for webhook authentication. The issue is fixed in version 2026.4.23, which reloads route secrets on each request, immediately invalidating rotated secrets without requiring restarts. The severity is medium, as exploitation requires prior possession of a valid secret.
OpenClaw CLI's gateway config mutation guard used a denylist to protect sensitive configuration paths, but the growing config schema outpaced this denylist, allowing unsafe model-driven config writes. This vulnerability enables a compromised or prompt-injected model with access to the owner-only gateway tool to persist unsafe configuration changes crossing security boundaries. These changes could affect command execution, network behavior, credentials, telemetry, and operator policies, and survive restarts. The issue affects openclaw versions from 2026.1.29 up to but not including 2026.4.23. The vulnerability is rated high severity with a CVSS score of 8.8. A fix is available in version 2026.4.23, which replaces the denylist with a fail-closed allowlist restricting config changes to narrow, safe paths.
OpenClaw CLI versions from 2026.1.29 up to but not including 2026.5.6 have a vulnerability where Telegram interactive callbacks can bypass the commands.allowFrom sender allowlist. This allows a Telegram user invoking an affected callback to mark the callback as authorized before the allowlist check is applied. The impact depends on the operator's configuration and whether untrusted input can reach the affected feature. A fix is available in version 2026.5.6. Until patched, restricting Telegram command callbacks to trusted chats and narrowing allowlists is recommended.
A vulnerability in the OpenClaw CLI Skill Workshop apply flow allows an agent tool call to override a pending approval state, applying changes prematurely. This affects versions from 2026.1.29 up to but not including 2026.5.6. The issue does not alter the trusted-operator model but could impact configurations where lower-trust inputs reach the affected path. The vulnerability has a medium severity with a CVSS score of 5.3.
A vulnerability in OpenClaw CLI's bundled MCP loopback allows bypassing an execution denylist during session spawn. This flaw enables broader command execution than intended when the affected feature is enabled and reachable. The issue affects versions from 2026.1.29 up to but not including 2026.5.12. The vulnerability does not alter OpenClaw's trusted-operator model but requires careful operator configuration to mitigate risk. A fix is available in version 2026.5.12.
OpenClaw CLI versions from 2026.1.29 up to but not including 2026.5.18 have a vulnerability in trusted-proxy Control UI mode where WebSocket clients could declare operator scopes before these were properly authorized. This allows an unpaired or restricted trusted-proxy client to gain operator.admin privileges on a live WebSocket connection, enabling admin-level Gateway RPCs until the connection ends or is revalidated. The issue does not affect shared-secret Control UI sessions. A fix is available in version 2026.5.18.
A vulnerability in OpenClaw CLI's Slack integration allows the 'allowFrom' policy to bind to mutable Slack display names. This means a Slack account that can change its display name metadata might match a policy entry incorrectly, potentially receiving agent access intended for another Slack identity. The issue affects versions from 2026.1.29 up to but not including 2026.5.4. The vulnerability does not alter OpenClaw's trusted-operator model. A fix is available starting with version 2026.5.3. Until patched, using stable Slack user IDs in allowlists and narrowing channel and tool allowlists are recommended mitigations.
A vulnerability in OpenClaw CLI's memory-wiki ingest feature allows a Gateway caller with operator.write access to read arbitrary local files instead of being restricted to intended ingest sources. This issue affects versions from 2026.1.29 up to but not including 2026.5.12. The vulnerability requires authenticated operator privileges and does not alter the trusted-operator security model of OpenClaw. The practical impact depends on the configuration and whether untrusted input can reach the vulnerable path. A patch is available to address this issue.