Security

Scope first: personal assistant security model

OpenClaw security guidance assumes a personal assistant deployment: one trusted operator boundary, potentially many agents.

• Supported security posture: one user/trust boundary per gateway (prefer one OS user/host/VPS per boundary).
• Not a supported security boundary: one shared gateway/agent used by mutually untrusted or adversarial users.
• If adversarial-user isolation is required, split by trust boundary (separate gateway + credentials, and ideally separate OS users/hosts).
• If multiple untrusted users can message one tool-enabled agent, treat them as sharing the same delegated tool authority for that agent.

This page explains hardening within that model. It does not claim hostile multi-tenant isolation on one shared gateway.

Quick check: openclaw security audit

See also: Formal Verification (Security Models)

Run this regularly (especially after changing config or exposing network surfaces):

openclaw security audit
openclaw security audit --deep
openclaw security audit --fix
openclaw security audit --json

security audit --fix stays intentionally narrow: it flips common open group policies to allowlists, restores logging.redactSensitive: "tools", tightens state/config/include-file permissions, and uses Windows ACL resets instead of POSIX chmod when running on Windows.

It flags common footguns (Gateway auth exposure, browser control exposure, elevated allowlists, filesystem permissions, permissive exec approvals, and open-channel tool exposure).

OpenClaw is both a product and an experiment: you’re wiring frontier-model behavior into real messaging surfaces and real tools. There is no “perfectly secure” setup. The goal is to be deliberate about:

• who can talk to your bot
• where the bot is allowed to act
• what the bot can touch

Start with the smallest access that still works, then widen it as you gain confidence.

Deployment and host trust

OpenClaw assumes the host and config boundary are trusted:

• If someone can modify Gateway host state/config (~/.openclaw, including openclaw.json), treat them as a trusted operator.
• Running one Gateway for multiple mutually untrusted/adversarial operators is not a recommended setup.
• For mixed-trust teams, split trust boundaries with separate gateways (or at minimum separate OS users/hosts).
• Recommended default: one user per machine/host (or VPS), one gateway for that user, and one or more agents in that gateway.
• Inside one Gateway instance, authenticated operator access is a trusted control-plane role, not a per-user tenant role.
• Session identifiers (sessionKey, session IDs, labels) are routing selectors, not authorization tokens.
• If several people can message one tool-enabled agent, each of them can steer that same permission set. Per-user session/memory isolation helps privacy, but does not convert a shared agent into per-user host authorization.

Shared Slack workspace: real risk

If "everyone in Slack can message the bot," the core risk is delegated tool authority:

• any allowed sender can induce tool calls (exec, browser, network/file tools) within the agent's policy;
• prompt/content injection from one sender can cause actions that affect shared state, devices, or outputs;
• if one shared agent has sensitive credentials/files, any allowed sender can potentially drive exfiltration via tool usage.

Use separate agents/gateways with minimal tools for team workflows; keep personal-data agents private.

Company-shared agent: acceptable pattern

This is acceptable when everyone using that agent is in the same trust boundary (for example one company team) and the agent is strictly business-scoped.

• run it on a dedicated machine/VM/container;
• use a dedicated OS user + dedicated browser/profile/accounts for that runtime;
• do not sign that runtime into personal Apple/Google accounts or personal password-manager/browser profiles.

If you mix personal and company identities on the same runtime, you collapse the separation and increase personal-data exposure risk.

Gateway and node trust concept

Treat Gateway and node as one operator trust domain, with different roles:

• Gateway is the control plane and policy surface (gateway.auth, tool policy, routing).
• Node is remote execution surface paired to that Gateway (commands, device actions, host-local capabilities).
• A caller authenticated to the Gateway is trusted at Gateway scope. After pairing, node actions are trusted operator actions on that node.
• Operator scope levels and approval-time checks are summarized in Operator scopes.
• Direct loopback backend clients authenticated with the shared gateway token/password can make internal control-plane RPCs without presenting a user device identity. This is not a remote or browser pairing bypass: network clients, node clients, device-token clients, and explicit device identities still go through pairing and scope-upgrade enforcement.
• sessionKey is routing/context selection, not per-user auth.
• Exec approvals (allowlist + ask) are guardrails for operator intent, not hostile multi-tenant isolation.
• OpenClaw's product default for trusted single-operator setups is that host exec on gateway/node is allowed without approval prompts (security="full", ask="off" unless you tighten it). That default is intentional UX, not a vulnerability by itself.
• Exec approvals bind exact request context and best-effort direct local file operands; they do not semantically model every runtime/interpreter loader path. Use sandboxing and host isolation for strong boundaries.

If you need hostile-user isolation, split trust boundaries by OS user/host and run separate gateways.

Trust boundary matrix

Use this as the quick model when triaging risk:

Not vulnerabilities by design

These patterns get reported often and are usually closed as no-action unless a real boundary bypass is demonstrated:

• Prompt-injection-only chains without a policy, auth, or sandbox bypass.
• Claims that assume hostile multi-tenant operation on one shared host or config.
• Claims that classify normal operator read-path access (for example sessions.list / sessions.preview / chat.history) as IDOR in a shared-gateway setup.
• Localhost-only deployment findings (for example HSTS on a loopback-only gateway).
• Discord inbound webhook signature findings for inbound paths that do not exist in this repo.
• Reports that treat node pairing metadata as a hidden second per-command approval layer for system.run, when the real execution boundary is still the gateway's global node command policy plus the node's own exec approvals.
• Reports that treat configured gateway.nodes.pairing.autoApproveCidrs as a vulnerability by itself. This setting is disabled by default, requires explicit CIDR/IP entries, only applies to first-time role: node pairing with no requested scopes, and does not auto-approve operator/browser/Control UI, WebChat, role upgrades, scope upgrades, metadata changes, public-key changes, or same-host loopback trusted-proxy header paths unless loopback trusted-proxy auth was explicitly enabled.
• "Missing per-user authorization" findings that treat sessionKey as an auth token.

Hardened baseline in 60 seconds

Use this baseline first, then selectively re-enable tools per trusted agent:

{
  gateway: {
    mode: "local",
    bind: "loopback",
    auth: { mode: "token", token: "replace-with-long-random-token" },
  },
  session: {
    dmScope: "per-channel-peer",
  },
  tools: {
    profile: "messaging",
    deny: ["group:automation", "group:runtime", "group:fs", "sessions_spawn", "sessions_send"],
    fs: { workspaceOnly: true },
    exec: { security: "deny", ask: "always" },
    elevated: { enabled: false },
  },
  channels: {
    whatsapp: { dmPolicy: "pairing", groups: { "*": { requireMention: true } } },
  },
}

This keeps the Gateway local-only, isolates DMs, and disables control-plane/runtime tools by default.

Shared inbox quick rule

If more than one person can DM your bot:

• Set session.dmScope: "per-channel-peer" (or "per-account-channel-peer" for multi-account channels).
• Keep dmPolicy: "pairing" or strict allowlists.
• Never combine shared DMs with broad tool access.
• This hardens cooperative/shared inboxes, but is not designed as hostile co-tenant isolation when users share host/config write access.

Context visibility model

OpenClaw separates two concepts:

• Trigger authorization: who can trigger the agent (dmPolicy, groupPolicy, allowlists, mention gates).
• Context visibility: what supplemental context is injected into model input (reply body, quoted text, thread history, forwarded metadata).

Allowlists gate triggers and command authorization. The contextVisibility setting controls how supplemental context (quoted replies, thread roots, fetched history) is filtered:

• contextVisibility: "all" (default) keeps supplemental context as received.
• contextVisibility: "allowlist" filters supplemental context to senders allowed by the active allowlist checks.
• contextVisibility: "allowlist_quote" behaves like allowlist, but still keeps one explicit quoted reply.

Set contextVisibility per channel or per room/conversation. See Group Chats for setup details.

Advisory triage guidance:

• Claims that only show "model can see quoted or historical text from non-allowlisted senders" are hardening findings addressable with contextVisibility, not auth or sandbox boundary bypasses by themselves.
• To be security-impacting, reports still need a demonstrated trust-boundary bypass (auth, policy, sandbox, approval, or another documented boundary).

What the audit checks (high level)

• Inbound access (DM policies, group policies, allowlists): can strangers trigger the bot?
• Tool blast radius (elevated tools + open rooms): could prompt injection turn into shell/file/network actions?
• Exec approval drift (security=full, autoAllowSkills, interpreter allowlists without strictInlineEval): are host-exec guardrails still doing what you think they are?
  • security="full" is a broad posture warning, not proof of a bug. It is the chosen default for trusted personal-assistant setups; tighten it only when your threat model needs approval or allowlist guardrails.
• Network exposure (Gateway bind/auth, Tailscale Serve/Funnel, weak/short auth tokens).
• Browser control exposure (remote nodes, relay ports, remote CDP endpoints).
• Local disk hygiene (permissions, symlinks, config includes, “synced folder” paths).
• Plugins (plugins load without an explicit allowlist).
• Policy drift/misconfig (sandbox docker settings configured but sandbox mode off; ineffective gateway.nodes.denyCommands patterns because matching is exact command-name only (for example system.run) and does not inspect shell text; dangerous gateway.nodes.allowCommands entries; global tools.profile="minimal" overridden by per-agent profiles; plugin-owned tools reachable under permissive tool policy).
• Runtime expectation drift (for example assuming implicit exec still means sandbox when tools.exec.host now defaults to auto, or explicitly setting tools.exec.host="sandbox" while sandbox mode is off).
• Model hygiene (warn when configured models look legacy; not a hard block).

If you run --deep, OpenClaw also attempts a best-effort live Gateway probe.

Credential storage map

Use this when auditing access or deciding what to back up:

• WhatsApp: ~/.openclaw/credentials/whatsapp/<accountId>/creds.json
• Telegram bot token: config/env or channels.telegram.tokenFile (regular file only; symlinks rejected)
• Discord bot token: config/env or SecretRef (env/file/exec providers)
• Slack tokens: config/env (channels.slack.*)
• Pairing allowlists:
  • ~/.openclaw/credentials/<channel>-allowFrom.json (default account)
  • ~/.openclaw/credentials/<channel>-<accountId>-allowFrom.json (non-default accounts)
• Model auth profiles: ~/.openclaw/agents/<agentId>/agent/auth-profiles.json
• Codex runtime state: ~/.openclaw/agents/<agentId>/agent/codex-home/
• File-backed secrets payload (optional): ~/.openclaw/secrets.json
• Legacy OAuth import: ~/.openclaw/credentials/oauth.json

Security audit checklist

When the audit prints findings, treat this as a priority order:

1. Anything “open” + tools enabled: lock down DMs/groups first (pairing/allowlists), then tighten tool policy/sandboxing.
2. Public network exposure (LAN bind, Funnel, missing auth): fix immediately.
3. Browser control remote exposure: treat it like operator access (tailnet-only, pair nodes deliberately, avoid public exposure).
4. Permissions: make sure state/config/credentials/auth are not group/world-readable.
5. Plugins: only load what you explicitly trust.
6. Model choice: prefer modern, instruction-hardened models for any bot with tools.

Security audit glossary

Each audit finding is keyed by a structured checkId (for example gateway.bind_no_auth or tools.exec.security_full_configured). Common critical severity classes:

• fs.* — filesystem permissions on state, config, credentials, auth profiles.
• gateway.* — bind mode, auth, Tailscale, Control UI, trusted-proxy setup.
• hooks.*, browser.*, sandbox.*, tools.exec.* — per-surface hardening.
• plugins.*, skills.* — plugin/skill supply chain and scan findings.
• security.exposure.* — cross-cutting checks where access policy meets tool blast radius.

See the full catalog with severity levels, fix keys, and auto-fix support at Security audit checks.

Control UI over HTTP

The Control UI needs a secure context (HTTPS or localhost) to generate device identity. gateway.controlUi.allowInsecureAuth is a local compatibility toggle:

• On localhost, it allows Control UI auth without device identity when the page is loaded over non-secure HTTP.
• It does not bypass pairing checks.
• It does not relax remote (non-localhost) device identity requirements.

Prefer HTTPS (Tailscale Serve) or open the UI on 127.0.0.1.

For break-glass scenarios only, gateway.controlUi.dangerouslyDisableDeviceAuth disables device identity checks entirely. This is a severe security downgrade; keep it off unless you are actively debugging and can revert quickly.

Separate from those dangerous flags, successful gateway.auth.mode: "trusted-proxy" can admit operator Control UI sessions without device identity. That is an intentional auth-mode behavior, not an allowInsecureAuth shortcut, and it still does not extend to node-role Control UI sessions.

openclaw security audit warns when this setting is enabled.

Insecure or dangerous flags summary

openclaw security audit raises config.insecure_or_dangerous_flags when known insecure/dangerous debug switches are enabled. Keep these unset in production.

- `gateway.controlUi.dangerouslyAllowHostHeaderOriginFallback`
- `gateway.controlUi.dangerouslyDisableDeviceAuth`
- `browser.ssrfPolicy.dangerouslyAllowPrivateNetwork`

Channel name-matching (bundled and plugin channels; also available per
`accounts.<accountId>` where applicable):

- `channels.discord.dangerouslyAllowNameMatching`
- `channels.slack.dangerouslyAllowNameMatching`
- `channels.googlechat.dangerouslyAllowNameMatching`
- `channels.msteams.dangerouslyAllowNameMatching`
- `channels.synology-chat.dangerouslyAllowNameMatching` (plugin channel)
- `channels.synology-chat.dangerouslyAllowInheritedWebhookPath` (plugin channel)
- `channels.zalouser.dangerouslyAllowNameMatching` (plugin channel)
- `channels.irc.dangerouslyAllowNameMatching` (plugin channel)
- `channels.mattermost.dangerouslyAllowNameMatching` (plugin channel)

Network exposure:

- `channels.telegram.network.dangerouslyAllowPrivateNetwork` (also per account)

Sandbox Docker (defaults + per-agent):

- `agents.defaults.sandbox.docker.dangerouslyAllowReservedContainerTargets`
- `agents.defaults.sandbox.docker.dangerouslyAllowExternalBindSources`
- `agents.defaults.sandbox.docker.dangerouslyAllowContainerNamespaceJoin`

Reverse proxy configuration

If you run the Gateway behind a reverse proxy (nginx, Caddy, Traefik, etc.), configure gateway.trustedProxies for proper forwarded-client IP handling.

When the Gateway detects proxy headers from an address that is not in trustedProxies, it will not treat connections as local clients. If gateway auth is disabled, those connections are rejected. This prevents authentication bypass where proxied connections would otherwise appear to come from localhost and receive automatic trust.

gateway.trustedProxies also feeds gateway.auth.mode: "trusted-proxy", but that auth mode is stricter:

• trusted-proxy auth fails closed on loopback-source proxies by default
• same-host loopback reverse proxies can use gateway.trustedProxies for local-client detection and forwarded IP handling
• same-host loopback reverse proxies can satisfy gateway.auth.mode: "trusted-proxy" only when gateway.auth.trustedProxy.allowLoopback = true; otherwise use token/password auth

gateway:
  trustedProxies:
    - "10.0.0.1" # reverse proxy IP
  # Optional. Default false.
  # Only enable if your proxy cannot provide X-Forwarded-For.
  allowRealIpFallback: false
  auth:
    mode: password
    password: ${OPENCLAW_GATEWAY_PASSWORD}

When trustedProxies is configured, the Gateway uses X-Forwarded-For to determine the client IP. X-Real-IP is ignored by default unless gateway.allowRealIpFallback: true is explicitly set.

Trusted proxy headers do not make node device pairing automatically trusted. gateway.nodes.pairing.autoApproveCidrs is a separate, disabled-by-default operator policy. Even when enabled, loopback-source trusted-proxy header paths are excluded from node auto-approval because local callers can forge those headers, including when loopback trusted-proxy auth is explicitly enabled.

Good reverse proxy behavior (overwrite incoming forwarding headers):

proxy_set_header X-Forwarded-For $remote_addr;
proxy_set_header X-Real-IP $remote_addr;

Bad reverse proxy behavior (append/preserve untrusted forwarding headers):

proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;

HSTS and origin notes

• OpenClaw gateway is local/loopback first. If you terminate TLS at a reverse proxy, set HSTS on the proxy-facing HTTPS domain there.
• If the gateway itself terminates HTTPS, you can set gateway.http.securityHeaders.strictTransportSecurity to emit the HSTS header from OpenClaw responses.
• Detailed deployment guidance is in Trusted Proxy Auth.
• For non-loopback Control UI deployments, gateway.controlUi.allowedOrigins is required by default.
• gateway.controlUi.allowedOrigins: ["*"] is an explicit allow-all browser-origin policy, not a hardened default. Avoid it outside tightly controlled local testing.
• Browser-origin auth failures on loopback are still rate-limited even when the general loopback exemption is enabled, but the lockout key is scoped per normalized Origin value instead of one shared localhost bucket.
• gateway.controlUi.dangerouslyAllowHostHeaderOriginFallback=true enables Host-header origin fallback mode; treat it as a dangerous operator-selected policy.
• Treat DNS rebinding and proxy-host header behavior as deployment hardening concerns; keep trustedProxies tight and avoid exposing the gateway directly to the public internet.

Local session logs live on disk

OpenClaw stores session transcripts on disk under ~/.openclaw/agents/<agentId>/sessions/*.jsonl. This is required for session continuity and (optionally) session memory indexing, but it also means any process/user with filesystem access can read those logs. Treat disk access as the trust boundary and lock down permissions on ~/.openclaw (see the audit section below). If you need stronger isolation between agents, run them under separate OS users or separate hosts.

Node execution (system.run)

If a macOS node is paired, the Gateway can invoke system.run on that node. This is remote code execution on the Mac:

• Requires node pairing (approval + token).
• Gateway node pairing is not a per-command approval surface. It establishes node identity/trust and token issuance.
• The Gateway applies a coarse global node command policy via gateway.nodes.allowCommands / denyCommands.
• Controlled on the Mac via Settings → Exec approvals (security + ask + allowlist).
• The per-node system.run policy is the node's own exec approvals file (exec.approvals.node.*), which can be stricter or looser than the gateway's global command-ID policy.
• A node running with security="full" and ask="off" is following the default trusted-operator model. Treat that as expected behavior unless your deployment explicitly requires a tighter approval or allowlist stance.
• Approval mode binds exact request context and, when possible, one concrete local script/file operand. If OpenClaw cannot identify exactly one direct local file for an interpreter/runtime command, approval-backed execution is denied rather than promising full semantic coverage.
• For host=node, approval-backed runs also store a canonical prepared systemRunPlan; later approved forwards reuse that stored plan, and gateway validation rejects caller edits to command/cwd/session context after the approval request was created.
• If you don’t want remote execution, set security to deny and remove node pairing for that Mac.

This distinction matters for triage:

• A reconnecting paired node advertising a different command list is not, by itself, a vulnerability if the Gateway global policy and the node's local exec approvals still enforce the actual execution boundary.
• Reports that treat node pairing metadata as a second hidden per-command approval layer are usually policy/UX confusion, not a security boundary bypass.

Dynamic skills (watcher / remote nodes)

OpenClaw can refresh the skills list mid-session:

• Skills watcher: changes to SKILL.md can update the skills snapshot on the next agent turn.
• Remote nodes: connecting a macOS node can make macOS-only skills eligible (based on bin probing).

Treat skill folders as trusted code and restrict who can modify them.

The threat model

Your AI assistant can:

• Execute arbitrary shell commands
• Read/write files
• Access network services
• Send messages to anyone (if you give it WhatsApp access)

People who message you can:

• Try to trick your AI into doing bad things
• Social engineer access to your data
• Probe for infrastructure details

Core concept: access control before intelligence

Most failures here are not fancy exploits — they’re “someone messaged the bot and the bot did what they asked.”

OpenClaw’s stance:

• Identity first: decide who can talk to the bot (DM pairing / allowlists / explicit “open”).
• Scope next: decide where the bot is allowed to act (group allowlists + mention gating, tools, sandboxing, device permissions).
• Model last: assume the model can be manipulated; design so manipulation has limited blast radius.

Command authorization model

Slash commands and directives are only honored for authorized senders. Authorization is derived from channel allowlists/pairing plus commands.useAccessGroups (see Configuration and Slash commands). If a channel allowlist is empty or includes "*", commands are effectively open for that channel.

/exec is a session-only convenience for authorized operators. It does not write config or change other sessions.

Control plane tools risk

Two built-in tools can make persistent control-plane changes:

• gateway can inspect config with config.schema.lookup / config.get, and can make persistent changes with config.apply, config.patch, and update.run.
• cron can create scheduled jobs that keep running after the original chat/task ends.

The owner-only gateway runtime tool still refuses to rewrite tools.exec.ask or tools.exec.security; legacy tools.bash.* aliases are normalized to the same protected exec paths before the write. Agent-driven gateway config.apply and gateway config.patch edits are fail-closed by default: only a narrow set of prompt, model, and mention-gating paths are agent-tunable. New sensitive config trees are therefore protected unless they are deliberately added to the allowlist.

For any agent/surface that handles untrusted content, deny these by default:

{
  tools: {
    deny: ["gateway", "cron", "sessions_spawn", "sessions_send"],
  },
}

commands.restart=false only blocks restart actions. It does not disable gateway config/update actions.

Plugins

Plugins run in-process with the Gateway. Treat them as trusted code:

• Only install plugins from sources you trust.
• Prefer explicit plugins.allow allowlists.
• Review plugin config before enabling.
• Restart the Gateway after plugin changes.
• If you install or update plugins (openclaw plugins install <package>, openclaw plugins update <id>), treat it like running untrusted code:
  • The install path is the per-plugin directory under the active plugin install root.
  • OpenClaw runs a built-in dangerous-code scan before install/update. critical findings block by default.
  • npm and git plugin installs run package-manager dependency convergence only during the explicit install/update flow. Local paths and archives are treated as self-contained plugin packages; OpenClaw copies/references them without running npm install.
  • Prefer pinned, exact versions (@scope/[email protected]), and inspect the unpacked code on disk before enabling.
  • --dangerously-force-unsafe-install is break-glass only for built-in scan false positives on plugin install/update flows. It does not bypass plugin before_install hook policy blocks and does not bypass scan failures.
  • Gateway-backed skill dependency installs follow the same dangerous/suspicious split: built-in critical findings block unless the caller explicitly sets dangerouslyForceUnsafeInstall, while suspicious findings still warn only. openclaw skills install remains the separate ClawHub skill download/install flow.

Details: Plugins

DM access model: pairing, allowlist, open, disabled

All current DM-capable channels support a DM policy (dmPolicy or *.dm.policy) that gates inbound DMs before the message is processed:

• pairing (default): unknown senders receive a short pairing code and the bot ignores their message until approved. Codes expire after 1 hour; repeated DMs won’t resend a code until a new request is created. Pending requests are capped at 3 per channel by default.
• allowlist: unknown senders are blocked (no pairing handshake).
• open: allow anyone to DM (public). Requires the channel allowlist to include "*" (explicit opt-in).
• disabled: ignore inbound DMs entirely.

Approve via CLI:

openclaw pairing list <channel>
openclaw pairing approve <channel> <code>

Details + files on disk: Pairing

DM session isolation (multi-user mode)

By default, OpenClaw routes all DMs into the main session so your assistant has continuity across devices and channels. If multiple people can DM the bot (open DMs or a multi-person allowlist), consider isolating DM sessions:

{
  session: { dmScope: "per-channel-peer" },
}

This prevents cross-user context leakage while keeping group chats isolated.

This is a messaging-context boundary, not a host-admin boundary. If users are mutually adversarial and share the same Gateway host/config, run separate gateways per trust boundary instead.

Secure DM mode (recommended)

Treat the snippet above as secure DM mode:

• Default: session.dmScope: "main" (all DMs share one session for continuity).
• Local CLI onboarding default: writes session.dmScope: "per-channel-peer" when unset (keeps existing explicit values).
• Secure DM mode: session.dmScope: "per-channel-peer" (each channel+sender pair gets an isolated DM context).
• Cross-channel peer isolation: session.dmScope: "per-peer" (each sender gets one session across all channels of the same type).

If you run multiple accounts on the same channel, use per-account-channel-peer instead. If the same person contacts you on multiple channels, use session.identityLinks to collapse those DM sessions into one canonical identity. See Session Management and Configuration.

Allowlists for DMs and groups

OpenClaw has two separate “who can trigger me?” layers:

• DM allowlist (allowFrom / channels.discord.allowFrom / channels.slack.allowFrom; legacy: channels.discord.dm.allowFrom, channels.slack.dm.allowFrom): who is allowed to talk to the bot in direct messages.
  • When dmPolicy="pairing", approvals are written to the account-scoped pairing allowlist store under ~/.openclaw/credentials/ (<channel>-allowFrom.json for default account, <channel>-<accountId>-allowFrom.json for non-default accounts), merged with config allowlists.
• Group allowlist (channel-specific): which groups/channels/guilds the bot will accept messages from at all.
  • Common patterns:
    • channels.whatsapp.groups, channels.telegram.groups, channels.imessage.groups: per-group defaults like requireMention; when set, it also acts as a group allowlist (include "*" to keep allow-all behavior).
    • groupPolicy="allowlist" + groupAllowFrom: restrict who can trigger the bot inside a group session (WhatsApp/Telegram/Signal/iMessage/Microsoft Teams).
    • channels.discord.guilds / channels.slack.channels: per-surface allowlists + mention defaults.
  • Group checks run in this order: groupPolicy/group allowlists first, mention/reply activation second.
  • Replying to a bot message (implicit mention) does not bypass sender allowlists like groupAllowFrom.
  • Security note: treat dmPolicy="open" and groupPolicy="open" as last-resort settings. They should be barely used; prefer pairing + allowlists unless you fully trust every member of the room.

Details: Configuration and Groups

Prompt injection (what it is, why it matters)

Prompt injection is when an attacker crafts a message that manipulates the model into doing something unsafe (“ignore your instructions”, “dump your filesystem”, “follow this link and run commands”, etc.).

Even with strong system prompts, prompt injection is not solved. System prompt guardrails are soft guidance only; hard enforcement comes from tool policy, exec approvals, sandboxing, and channel allowlists (and operators can disable these by design). What helps in practice:

• Keep inbound DMs locked down (pairing/allowlists).
• Prefer mention gating in groups; avoid “always-on” bots in public rooms.
• Treat links, attachments, and pasted instructions as hostile by default.
• Run sensitive tool execution in a sandbox; keep secrets out of the agent’s reachable filesystem.
• Note: sandboxing is opt-in. If sandbox mode is off, implicit host=auto resolves to the gateway host. Explicit host=sandbox still fails closed because no sandbox runtime is available. Set host=gateway if you want that behavior to be explicit in config.
• Limit high-risk tools (exec, browser, web_fetch, web_search) to trusted agents or explicit allowlists.
• If you allowlist interpreters (python, node, ruby, perl, php, lua, osascript), enable tools.exec.strictInlineEval so inline eval forms still need explicit approval.
• Shell approval analysis also rejects POSIX parameter-expansion forms ($VAR, $?, $$, $1, $@, ${…}) inside unquoted heredocs, so an allowlisted heredoc body cannot sneak shell expansion past allowlist review as plain text. Quote the heredoc terminator (for example <<'EOF') to opt into literal body semantics; unquoted heredocs that would have expanded variables are rejected.
• Model choice matters: older/smaller/legacy models are significantly less robust against prompt injection and tool misuse. For tool-enabled agents, use the strongest latest-generation, instruction-hardened model available.

Red flags to treat as untrusted:

• “Read this file/URL and do exactly what it says.”
• “Ignore your system prompt or safety rules.”
• “Reveal your hidden instructions or tool outputs.”
• “Paste the full contents of ~/.openclaw or your logs.”

External content special-token sanitization

OpenClaw strips common self-hosted LLM chat-template special-token literals from wrapped external content and metadata before they reach the model. Covered marker families include Qwen/ChatML, Llama, Gemma, Mistral, Phi, and GPT-OSS role/turn tokens.

Why:

• OpenAI-compatible backends that front self-hosted models sometimes preserve special tokens that appear in user text, instead of masking them. An attacker who can write into inbound external content (a fetched page, an email body, a file contents tool output) could otherwise inject a synthetic assistant or system role boundary and escape the wrapped-content guardrails.
• Sanitization happens at the external-content wrapping layer, so it applies uniformly across fetch/read tools and inbound channel content rather than being per-provider.
• Outbound model responses already have a separate sanitizer that strips leaked <tool_call>, <function_calls>, ``, <previous_response>, and similar internal runtime scaffolding from user-visible replies at the final channel delivery boundary. The external-content sanitizer is the inbound counterpart.

This does not replace the other hardening on this page — dmPolicy, allowlists, exec approvals, sandboxing, and contextVisibility still do the primary work. It closes one specific tokenizer-layer bypass against self-hosted stacks that forward user text with special tokens intact.

Unsafe external content bypass flags

OpenClaw includes explicit bypass flags that disable external-content safety wrapping:

• hooks.mappings[].allowUnsafeExternalContent
• hooks.gmail.allowUnsafeExternalContent
• Cron payload field allowUnsafeExternalContent

Guidance:

• Keep these unset/false in production.
• Only enable temporarily for tightly scoped debugging.
• If enabled, isolate that agent (sandbox + minimal tools + dedicated session namespace).

Hooks risk note:

• Hook payloads are untrusted content, even when delivery comes from systems you control (mail/docs/web content can carry prompt injection).
• Weak model tiers increase this risk. For hook-driven automation, prefer strong modern model tiers and keep tool policy tight (tools.profile: "messaging" or stricter), plus sandboxing where possible.

Prompt injection does not require public DMs

Even if only you can message the bot, prompt injection can still happen via any untrusted content the bot reads (web search/fetch results, browser pages, emails, docs, attachments, pasted logs/code). In other words: the sender is not the only threat surface; the content itself can carry adversarial instructions.

When tools are enabled, the typical risk is exfiltrating context or triggering tool calls. Reduce the blast radius by:

• Using a read-only or tool-disabled reader agent to summarize untrusted content, then pass the summary to your main agent.
• Keeping web_search / web_fetch / browser off for tool-enabled agents unless needed.
• For OpenResponses URL inputs (input_file / input_image), set tight gateway.http.endpoints.responses.files.urlAllowlist and gateway.http.endpoints.responses.images.urlAllowlist, and keep maxUrlParts low. Empty allowlists are treated as unset; use files.allowUrl: false / images.allowUrl: false if you want to disable URL fetching entirely.
• For OpenResponses file inputs, decoded input_file text is still injected as untrusted external content. Do not rely on file text being trusted just because the Gateway decoded it locally. The injected block still carries explicit <<<EXTERNAL_UNTRUSTED_CONTENT ...>>> boundary markers plus Source: External metadata, even though this path omits the longer SECURITY NOTICE: banner.
• The same marker-based wrapping is applied when media-understanding extracts text from attached documents before appending that text to the media prompt.
• Enabling sandboxing and strict tool allowlists for any agent that touches untrusted input.
• Keeping secrets out of prompts; pass them via env/config on the gateway host instead.

Self-hosted LLM backends

OpenAI-compatible self-hosted backends such as vLLM, SGLang, TGI, LM Studio, or custom Hugging Face tokenizer stacks can differ from hosted providers in how chat-template special tokens are handled. If a backend tokenizes literal strings such as <|im_start|>, <|start_header_id|>, or <start_of_turn> as structural chat-template tokens inside user content, untrusted text can try to forge role boundaries at the tokenizer layer.

OpenClaw strips common model-family special-token literals from wrapped external content before dispatching it to the model. Keep external-content wrapping enabled, and prefer backend settings that split or escape special tokens in user-provided content when available. Hosted providers such as OpenAI and Anthropic already apply their own request-side sanitization.

Model strength (security note)

Prompt injection resistance is not uniform across model tiers. Smaller/cheaper models are generally more susceptible to tool misuse and instruction hijacking, especially under adversarial prompts.

Recommendations:

• Use the latest generation, best-tier model for any bot that can run tools or touch files/networks.
• Do not use older/weaker/smaller tiers for tool-enabled agents or untrusted inboxes; the prompt-injection risk is too high.
• If you must use a smaller model, reduce blast radius (read-only tools, strong sandboxing, minimal filesystem access, strict allowlists).
• When running small models, enable sandboxing for all sessions and disable web_search/web_fetch/browser unless inputs are tightly controlled.
• For chat-only personal assistants with trusted input and no tools, smaller models are usually fine.

Reasoning and verbose output in groups

/reasoning, /verbose, and /trace can expose internal reasoning, tool output, or plugin diagnostics that was not meant for a public channel. In group settings, treat them as debug only and keep them off unless you explicitly need them.

Guidance:

• Keep /reasoning, /verbose, and /trace disabled in public rooms.
• If you enable them, do so only in trusted DMs or tightly controlled rooms.
• Remember: verbose and trace output can include tool args, URLs, plugin diagnostics, and data the model saw.

Configuration hardening examples

File permissions

Keep config + state private on the gateway host:

• ~/.openclaw/openclaw.json: 600 (user read/write only)
• ~/.openclaw: 700 (user only)

openclaw doctor can warn and offer to tighten these permissions.

Network exposure (bind, port, firewall)

The Gateway multiplexes WebSocket + HTTP on a single port:

• Default: 18789
• Config/flags/env: gateway.port, --port, OPENCLAW_GATEWAY_PORT

This HTTP surface includes the Control UI and the canvas host:

• Control UI (SPA assets) (default base path /)
• Canvas host: /__openclaw__/canvas/ and /__openclaw__/a2ui/ (arbitrary HTML/JS; treat as untrusted content)

If you load canvas content in a normal browser, treat it like any other untrusted web page:

• Don't expose the canvas host to untrusted networks/users.
• Don't make canvas content share the same origin as privileged web surfaces unless you fully understand the implications.

Bind mode controls where the Gateway listens:

• gateway.bind: "loopback" (default): only local clients can connect.
• Non-loopback binds ("lan", "tailnet", "custom") expand the attack surface. Only use them with gateway auth (shared token/password or a correctly configured trusted proxy) and a real firewall.

Rules of thumb:

• Prefer Tailscale Serve over LAN binds (Serve keeps the Gateway on loopback, and Tailscale handles access).
• If you must bind to LAN, firewall the port to a tight allowlist of source IPs; do not port-forward it broadly.
• Never expose the Gateway unauthenticated on 0.0.0.0.

Docker port publishing with UFW

If you run OpenClaw with Docker on a VPS, remember that published container ports (-p HOST:CONTAINER or Compose ports:) are routed through Docker's forwarding chains, not only host INPUT rules.

To keep Docker traffic aligned with your firewall policy, enforce rules in DOCKER-USER (this chain is evaluated before Docker's own accept rules). On many modern distros, iptables/ip6tables use the iptables-nft frontend and still apply these rules to the nftables backend.

Minimal allowlist example (IPv4):

# /etc/ufw/after.rules (append as its own *filter section)
*filter
:DOCKER-USER - [0:0]
-A DOCKER-USER -m conntrack --ctstate ESTABLISHED,RELATED -j RETURN
-A DOCKER-USER -s 127.0.0.0/8 -j RETURN
-A DOCKER-USER -s 10.0.0.0/8 -j RETURN
-A DOCKER-USER -s 172.16.0.0/12 -j RETURN
-A DOCKER-USER -s 192.168.0.0/16 -j RETURN
-A DOCKER-USER -s 100.64.0.0/10 -j RETURN
-A DOCKER-USER -p tcp --dport 80 -j RETURN
-A DOCKER-USER -p tcp --dport 443 -j RETURN
-A DOCKER-USER -m conntrack --ctstate NEW -j DROP
-A DOCKER-USER -j RETURN
COMMIT

IPv6 has separate tables. Add a matching policy in /etc/ufw/after6.rules if Docker IPv6 is enabled.

Avoid hardcoding interface names like eth0 in docs snippets. Interface names vary across VPS images (ens3, enp*, etc.) and mismatches can accidentally skip your deny rule.

Quick validation after reload:

ufw reload
iptables -S DOCKER-USER
ip6tables -S DOCKER-USER
nmap -sT -p 1-65535 <public-ip> --open

Expected external ports should be only what you intentionally expose (for most setups: SSH + your reverse proxy ports).

mDNS/Bonjour discovery

When the bundled bonjour plugin is enabled, the Gateway broadcasts its presence via mDNS (_openclaw-gw._tcp on port 5353) for local device discovery. In full mode, this includes TXT records that may expose operational details:

• cliPath: full filesystem path to the CLI binary (reveals username and install location)
• sshPort: advertises SSH availability on the host
• displayName, lanHost: hostname information

Operational security consideration: Broadcasting infrastructure details makes reconnaissance easier for anyone on the local network. Even "harmless" info like filesystem paths and SSH availability helps attackers map your environment.

Recommendations:

1. Keep Bonjour disabled unless LAN discovery is needed. Bonjour auto-starts on macOS hosts and is opt-in elsewhere; direct Gateway URLs, Tailnet, SSH, or wide-area DNS-SD avoid local multicast.

2. Minimal mode (default when Bonjour is enabled, recommended for exposed gateways): omit sensitive fields from mDNS broadcasts:

   json5Copy

   {
     discovery: {
       mdns: { mode: "minimal" },
     },
   }
   

3. Disable mDNS mode if you want to keep the plugin enabled but suppress local device discovery:

   json5Copy

   {
     discovery: {
       mdns: { mode: "off" },
     },
   }
   

4. Full mode (opt-in): include cliPath + sshPort in TXT records:

   json5Copy

   {
     discovery: {
       mdns: { mode: "full" },
     },
   }
   

5. Environment variable (alternative): set OPENCLAW_DISABLE_BONJOUR=1 to disable mDNS without config changes.

When Bonjour is enabled in minimal mode, the Gateway broadcasts enough for device discovery (role, gatewayPort, transport) but omits cliPath and sshPort. Apps that need CLI path information can fetch it via the authenticated WebSocket connection instead.

Lock down the Gateway WebSocket (local auth)

Gateway auth is required by default. If no valid gateway auth path is configured, the Gateway refuses WebSocket connections (fail‑closed).

Onboarding generates a token by default (even for loopback) so local clients must authenticate.

Set a token so all WS clients must authenticate:

{
  gateway: {
    auth: { mode: "token", token: "your-token" },
  },
}

Doctor can generate one for you: openclaw doctor --generate-gateway-token.

Local device pairing:

• Device pairing is auto-approved for direct local loopback connects to keep same-host clients smooth.
• OpenClaw also has a narrow backend/container-local self-connect path for trusted shared-secret helper flows.
• Tailnet and LAN connects, including same-host tailnet binds, are treated as remote for pairing and still need approval.
• Forwarded-header evidence on a loopback request disqualifies loopback locality. Metadata-upgrade auto-approval is scoped narrowly. See Gateway pairing for both rules.

Auth modes:

• gateway.auth.mode: "token": shared bearer token (recommended for most setups).
• gateway.auth.mode: "password": password auth (prefer setting via env: OPENCLAW_GATEWAY_PASSWORD).
• gateway.auth.mode: "trusted-proxy": trust an identity-aware reverse proxy to authenticate users and pass identity via headers (see Trusted Proxy Auth).

Rotation checklist (token/password):

1. Generate/set a new secret (gateway.auth.token or OPENCLAW_GATEWAY_PASSWORD).
2. Restart the Gateway (or restart the macOS app if it supervises the Gateway).
3. Update any remote clients (gateway.remote.token / .password on machines that call into the Gateway).
4. Verify you can no longer connect with the old credentials.

Tailscale Serve identity headers

When gateway.auth.allowTailscale is true (default for Serve), OpenClaw accepts Tailscale Serve identity headers (tailscale-user-login) for Control UI/WebSocket authentication. OpenClaw verifies the identity by resolving the x-forwarded-for address through the local Tailscale daemon (tailscale whois) and matching it to the header. This only triggers for requests that hit loopback and include x-forwarded-for, x-forwarded-proto, and x-forwarded-host as injected by Tailscale. For this async identity check path, failed attempts for the same {scope, ip} are serialized before the limiter records the failure. Concurrent bad retries from one Serve client can therefore lock out the second attempt immediately instead of racing through as two plain mismatches. HTTP API endpoints (for example /v1/*, /tools/invoke, and /api/channels/*) do not use Tailscale identity-header auth. They still follow the gateway's configured HTTP auth mode.

Important boundary note:

• Gateway HTTP bearer auth is effectively all-or-nothing operator access.
• Treat credentials that can call /v1/chat/completions, /v1/responses, or /api/channels/* as full-access operator secrets for that gateway.
• On the OpenAI-compatible HTTP surface, shared-secret bearer auth restores the full default operator scopes (operator.admin, operator.approvals, operator.pairing, operator.read, operator.talk.secrets, operator.write) and owner semantics for agent turns; narrower x-openclaw-scopes values do not reduce that shared-secret path.
• Per-request scope semantics on HTTP only apply when the request comes from an identity-bearing mode such as trusted proxy auth or gateway.auth.mode="none" on a private ingress.
• In those identity-bearing modes, omitting x-openclaw-scopes falls back to the normal operator default scope set; send the header explicitly when you want a narrower scope set.
• /tools/invoke follows the same shared-secret rule: token/password bearer auth is treated as full operator access there too, while identity-bearing modes still honor declared scopes.
• Do not share these credentials with untrusted callers; prefer separate gateways per trust boundary.

Trust assumption: tokenless Serve auth assumes the gateway host is trusted. Do not treat this as protection against hostile same-host processes. If untrusted local code may run on the gateway host, disable gateway.auth.allowTailscale and require explicit shared-secret auth with gateway.auth.mode: "token" or "password".

Security rule: do not forward these headers from your own reverse proxy. If you terminate TLS or proxy in front of the gateway, disable gateway.auth.allowTailscale and use shared-secret auth (gateway.auth.mode: "token" or "password") or Trusted Proxy Auth instead.

Trusted proxies:

• If you terminate TLS in front of the Gateway, set gateway.trustedProxies to your proxy IPs.
• OpenClaw will trust x-forwarded-for (or x-real-ip) from those IPs to determine the client IP for local pairing checks and HTTP auth/local checks.
• Ensure your proxy overwrites x-forwarded-for and blocks direct access to the Gateway port.

See Tailscale and Web overview.

Browser control via node host (recommended)

If your Gateway is remote but the browser runs on another machine, run a node host on the browser machine and let the Gateway proxy browser actions (see Browser tool). Treat node pairing like admin access.

Recommended pattern:

• Keep the Gateway and node host on the same tailnet (Tailscale).
• Pair the node intentionally; disable browser proxy routing if you don’t need it.

Avoid:

• Exposing relay/control ports over LAN or public Internet.
• Tailscale Funnel for browser control endpoints (public exposure).

Secrets on disk

Assume anything under ~/.openclaw/ (or $OPENCLAW_STATE_DIR/) may contain secrets or private data:

• openclaw.json: config may include tokens (gateway, remote gateway), provider settings, and allowlists.
• credentials/**: channel credentials (example: WhatsApp creds), pairing allowlists, legacy OAuth imports.
• agents/<agentId>/agent/auth-profiles.json: API keys, token profiles, OAuth tokens, and optional keyRef/tokenRef.
• agents/<agentId>/agent/codex-home/**: per-agent Codex app-server account, config, skills, plugins, native thread state, and diagnostics.
• secrets.json (optional): file-backed secret payload used by file SecretRef providers (secrets.providers).
• agents/<agentId>/agent/auth.json: legacy compatibility file. Static api_key entries are scrubbed when discovered.
• agents/<agentId>/sessions/**: session transcripts (*.jsonl) + routing metadata (sessions.json) that can contain private messages and tool output.
• bundled plugin packages: installed plugins (plus their node_modules/).
• sandboxes/**: tool sandbox workspaces; can accumulate copies of files you read/write inside the sandbox.

Hardening tips:

• Keep permissions tight (700 on dirs, 600 on files).
• Use full-disk encryption on the gateway host.
• Prefer a dedicated OS user account for the Gateway if the host is shared.

Workspace .env files

OpenClaw loads workspace-local .env files for agents and tools, but never lets those files silently override gateway runtime controls.

• Any key that starts with OPENCLAW_* is blocked from untrusted workspace .env files.
• Channel endpoint settings for Matrix, Mattermost, IRC, and Synology Chat are also blocked from workspace .env overrides, so cloned workspaces cannot redirect bundled connector traffic through local endpoint config. Endpoint env keys (such as MATRIX_HOMESERVER, MATTERMOST_URL, IRC_HOST, SYNOLOGY_CHAT_INCOMING_URL) must come from the gateway process environment or env.shellEnv, not from a workspace-loaded .env.
• The block is fail-closed: a new runtime-control variable added in a future release cannot be inherited from a checked-in or attacker-supplied .env; the key is ignored and the gateway keeps its own value.
• Trusted process/OS environment variables (the gateway's own shell, launchd/systemd unit, app bundle) still apply — this only constrains .env file loading.

Why: workspace .env files frequently live next to agent code, get committed by accident, or get written by tools. Blocking the whole OPENCLAW_* prefix means adding a new OPENCLAW_* flag later can never regress into silent inheritance from workspace state.

Logs and transcripts (redaction and retention)

Logs and transcripts can leak sensitive info even when access controls are correct:

• Gateway logs may include tool summaries, errors, and URLs.
• Session transcripts can include pasted secrets, file contents, command output, and links.

Recommendations:

• Keep log and transcript redaction on (logging.redactSensitive: "tools"; default).
• Add custom patterns for your environment via logging.redactPatterns (tokens, hostnames, internal URLs).
• When sharing diagnostics, prefer openclaw status --all (pasteable, secrets redacted) over raw logs.
• Prune old session transcripts and log files if you don’t need long retention.

Details: Logging

DMs: pairing by default

{
  channels: { whatsapp: { dmPolicy: "pairing" } },
}

Groups: require mention everywhere

{
  "channels": {
    "whatsapp": {
      "groups": {
        "*": { "requireMention": true }
      }
    }
  },
  "agents": {
    "list": [
      {
        "id": "main",
        "groupChat": { "mentionPatterns": ["@openclaw", "@mybot"] }
      }
    ]
  }
}

In group chats, only respond when explicitly mentioned.

Separate numbers (WhatsApp, Signal, Telegram)

For phone-number-based channels, consider running your AI on a separate phone number from your personal one:

• Personal number: Your conversations stay private
• Bot number: AI handles these, with appropriate boundaries

Read-only mode (via sandbox and tools)

You can build a read-only profile by combining:

• agents.defaults.sandbox.workspaceAccess: "ro" (or "none" for no workspace access)
• tool allow/deny lists that block write, edit, apply_patch, exec, process, etc.

Additional hardening options:

• tools.exec.applyPatch.workspaceOnly: true (default): ensures apply_patch cannot write/delete outside the workspace directory even when sandboxing is off. Set to false only if you intentionally want apply_patch to touch files outside the workspace.
• tools.fs.workspaceOnly: true (optional): restricts read/write/edit/apply_patch paths and native prompt image auto-load paths to the workspace directory (useful if you allow absolute paths today and want a single guardrail).
• Keep filesystem roots narrow: avoid broad roots like your home directory for agent workspaces/sandbox workspaces. Broad roots can expose sensitive local files (for example state/config under ~/.openclaw) to filesystem tools.

Secure baseline (copy/paste)

One “safe default” config that keeps the Gateway private, requires DM pairing, and avoids always-on group bots:

{
  gateway: {
    mode: "local",
    bind: "loopback",
    port: 18789,
    auth: { mode: "token", token: "your-long-random-token" },
  },
  channels: {
    whatsapp: {
      dmPolicy: "pairing",
      groups: { "*": { requireMention: true } },
    },
  },
}

If you want “safer by default” tool execution too, add a sandbox + deny dangerous tools for any non-owner agent (example below under “Per-agent access profiles”).

Built-in baseline for chat-driven agent turns: non-owner senders cannot use the cron or gateway tools.

Sandboxing (recommended)

Dedicated doc: Sandboxing

Two complementary approaches:

• Run the full Gateway in Docker (container boundary): Docker
• Tool sandbox (agents.defaults.sandbox, host gateway + sandbox-isolated tools; Docker is the default backend): Sandboxing

Also consider agent workspace access inside the sandbox:

• agents.defaults.sandbox.workspaceAccess: "none" (default) keeps the agent workspace off-limits; tools run against a sandbox workspace under ~/.openclaw/sandboxes
• agents.defaults.sandbox.workspaceAccess: "ro" mounts the agent workspace read-only at /agent (disables write/edit/apply_patch)
• agents.defaults.sandbox.workspaceAccess: "rw" mounts the agent workspace read/write at /workspace
• Extra sandbox.docker.binds are validated against normalized and canonicalized source paths. Parent-symlink tricks and canonical home aliases still fail closed if they resolve into blocked roots such as /etc, /var/run, or credential directories under the OS home.

Sub-agent delegation guardrail

If you allow session tools, treat delegated sub-agent runs as another boundary decision:

• Deny sessions_spawn unless the agent truly needs delegation.
• Keep agents.defaults.subagents.allowAgents and any per-agent agents.list[].subagents.allowAgents overrides restricted to known-safe target agents.
• For any workflow that must remain sandboxed, call sessions_spawn with sandbox: "require" (default is inherit).
• sandbox: "require" fails fast when the target child runtime is not sandboxed.

Browser control risks

Enabling browser control gives the model the ability to drive a real browser. If that browser profile already contains logged-in sessions, the model can access those accounts and data. Treat browser profiles as sensitive state:

• Prefer a dedicated profile for the agent (the default openclaw profile).
• Avoid pointing the agent at your personal daily-driver profile.
• Keep host browser control disabled for sandboxed agents unless you trust them.
• The standalone loopback browser control API only honors shared-secret auth (gateway token bearer auth or gateway password). It does not consume trusted-proxy or Tailscale Serve identity headers.
• Treat browser downloads as untrusted input; prefer an isolated downloads directory.
• Disable browser sync/password managers in the agent profile if possible (reduces blast radius).
• For remote gateways, assume “browser control” is equivalent to “operator access” to whatever that profile can reach.
• Keep the Gateway and node hosts tailnet-only; avoid exposing browser control ports to LAN or public Internet.
• Disable browser proxy routing when you don’t need it (gateway.nodes.browser.mode="off").
• Chrome MCP existing-session mode is not “safer”; it can act as you in whatever that host Chrome profile can reach.

Browser SSRF policy (strict by default)

OpenClaw’s browser navigation policy is strict by default: private/internal destinations stay blocked unless you explicitly opt in.

• Default: browser.ssrfPolicy.dangerouslyAllowPrivateNetwork is unset, so browser navigation keeps private/internal/special-use destinations blocked.
• Legacy alias: browser.ssrfPolicy.allowPrivateNetwork is still accepted for compatibility.
• Opt-in mode: set browser.ssrfPolicy.dangerouslyAllowPrivateNetwork: true to allow private/internal/special-use destinations.
• In strict mode, use hostnameAllowlist (patterns like *.example.com) and allowedHostnames (exact host exceptions, including blocked names like localhost) for explicit exceptions.
• Navigation is checked before request and best-effort re-checked on the final http(s) URL after navigation to reduce redirect-based pivots.

Example strict policy:

{
  browser: {
    ssrfPolicy: {
      dangerouslyAllowPrivateNetwork: false,
      hostnameAllowlist: ["*.example.com", "example.com"],
      allowedHostnames: ["localhost"],
    },
  },
}

Per-agent access profiles (multi-agent)

With multi-agent routing, each agent can have its own sandbox + tool policy: use this to give full access, read-only, or no access per agent. See Multi-Agent Sandbox & Tools for full details and precedence rules.

Common use cases:

• Personal agent: full access, no sandbox
• Family/work agent: sandboxed + read-only tools
• Public agent: sandboxed + no filesystem/shell tools

Example: full access (no sandbox)

{
  agents: {
    list: [
      {
        id: "personal",
        workspace: "~/.openclaw/workspace-personal",
        sandbox: { mode: "off" },
      },
    ],
  },
}

Example: read-only tools + read-only workspace

{
  agents: {
    list: [
      {
        id: "family",
        workspace: "~/.openclaw/workspace-family",
        sandbox: {
          mode: "all",
          scope: "agent",
          workspaceAccess: "ro",
        },
        tools: {
          allow: ["read"],
          deny: ["write", "edit", "apply_patch", "exec", "process", "browser"],
        },
      },
    ],
  },
}

Example: no filesystem/shell access (provider messaging allowed)

{
  agents: {
    list: [
      {
        id: "public",
        workspace: "~/.openclaw/workspace-public",
        sandbox: {
          mode: "all",
          scope: "agent",
          workspaceAccess: "none",
        },
        // Session tools can reveal sensitive data from transcripts. By default OpenClaw limits these tools
        // to the current session + spawned subagent sessions, but you can clamp further if needed.
        // See `tools.sessions.visibility` in the configuration reference.
        tools: {
          sessions: { visibility: "tree" }, // self | tree | agent | all
          allow: [
            "sessions_list",
            "sessions_history",
            "sessions_send",
            "sessions_spawn",
            "session_status",
            "whatsapp",
            "telegram",
            "slack",
            "discord",
          ],
          deny: [
            "read",
            "write",
            "edit",
            "apply_patch",
            "exec",
            "process",
            "browser",
            "canvas",
            "nodes",
            "cron",
            "gateway",
            "image",
          ],
        },
      },
    ],
  },
}

Incident response

If your AI does something bad:

Contain

1. Stop it: stop the macOS app (if it supervises the Gateway) or terminate your openclaw gateway process.
2. Close exposure: set gateway.bind: "loopback" (or disable Tailscale Funnel/Serve) until you understand what happened.
3. Freeze access: switch risky DMs/groups to dmPolicy: "disabled" / require mentions, and remove "*" allow-all entries if you had them.

Rotate (assume compromise if secrets leaked)

1. Rotate Gateway auth (gateway.auth.token / OPENCLAW_GATEWAY_PASSWORD) and restart.
2. Rotate remote client secrets (gateway.remote.token / .password) on any machine that can call the Gateway.
3. Rotate provider/API credentials (WhatsApp creds, Slack/Discord tokens, model/API keys in auth-profiles.json, and encrypted secrets payload values when used).

Audit

1. Check Gateway logs: /tmp/openclaw/openclaw-YYYY-MM-DD.log (or logging.file).
2. Review the relevant transcript(s): ~/.openclaw/agents/<agentId>/sessions/*.jsonl.
3. Review recent config changes (anything that could have widened access: gateway.bind, gateway.auth, dm/group policies, tools.elevated, plugin changes).
4. Re-run openclaw security audit --deep and confirm critical findings are resolved.

Collect for a report

• Timestamp, gateway host OS + OpenClaw version
• The session transcript(s) + a short log tail (after redacting)
• What the attacker sent + what the agent did
• Whether the Gateway was exposed beyond loopback (LAN/Tailscale Funnel/Serve)

Secret scanning

CI runs the pre-commit detect-private-key hook over the repository. If it fails, remove or rotate the committed key material, then reproduce locally:

pre-commit run --all-files detect-private-key

Reporting security issues

Found a vulnerability in OpenClaw? Please report responsibly:

1. Email: [email protected]
2. Don't post publicly until fixed
3. We'll credit you (unless you prefer anonymity)