OpenClaw security.exposure.open_groups_with_elevated: What It Means and How to Fix It
Published 2026-10-20 · Updated 2026-10-20 · By The Is My Agent Safe team
Short answer
At least one chat channel lets anyone trigger your agent (a group or DM policy set to open), and elevated mode isn't turned off. Fix the open policy first: switch groups to allowlist and DMs to pairing or allowlist. If you don't use elevated mode, set tools.elevated.enabled to false. Then run openclaw security audit --deep again.
- OpenClaw rates it critical. In 2026.10.5's source, it fires when a channel or channel account has groupPolicy or dmPolicy set to "open" and tools.elevated.enabled isn't false.
- Fix the open policy first. OpenClaw's audit triage order puts anything "open" with tools enabled at the top.
- openclaw security audit --fix switches open group policies to allowlist. Its core fix code doesn't change DM policies, so check those yourself.
- An open GitHub report (#175215) says this check can fire when the open channel is bound only to a restricted agent. OpenClaw's maintainers haven't confirmed it or shipped a fix.
Paste your audit: run openclaw security audit --deep and paste the output into our free check for a red/amber/green result with the most important finding first (Limits of the check).
What should I do first?
The finding's detail lists every policy set to "open", such as channels.telegram.groupPolicy or channels.slack.accounts.work.dmPolicy. Close each one, then decide whether you need elevated mode at all:
openclaw gateway stop openclaw config set channels.<channel>.groupPolicy allowlist # for each groupPolicy in the detail openclaw config set channels.<channel>.dmPolicy pairing # for each dmPolicy in the detail openclaw config set tools.elevated.enabled false # only if you don't use elevated mode openclaw config validate openclaw gateway start openclaw security audit --deep
Use the exact paths from the detail, including any accounts.<account> part. OpenClaw's config docs say config set needs the Gateway stopped, which is why the commands stop it first. When the audit comes back without this finding, you're done with it.
OpenClaw's audit docs say openclaw security audit --fix "flips open group policies to allowlists". In 2026.10.5's fix code, the core change covers groupPolicy for each channel and channel account, not DM policies. Check the DM paths yourself after running it.
What changes for the people who message your agent?
- Groups on
allowlist: OpenClaw's groups docs say "if your group allowlist is empty, group messages are blocked". To let only you trigger the agent in groups, the docs showgroupPolicy: "allowlist"with your own ID ingroupAllowFrom. - DMs on
pairing: OpenClaw's pairing docs say "unknown senders get a short code and their message is not processed until you approve". Approve someone withopenclaw pairing approve <channel> <CODE>. Pairing works on channels that support OpenClaw's pairing API. On other channels, useallowlistwith the sender IDs you trust.
What does security.exposure.open_groups_with_elevated mean?
OpenClaw's security audit checks page lists it as critical: "Open DMs/groups + elevated tools create high-impact prompt-injection paths".
In OpenClaw 2026.10.5's source, the finding appears when both of these are true:
- At least one channel, or an account under a channel, has
groupPolicyordmPolicyset to"open". The olderdm.policysetting counts too. tools.elevated.enabledisn't set tofalse.
Made-up example of the detail line:
Found inbound policy="open" at: - channels.telegram.groupPolicy With tools.elevated enabled, a prompt injection in those conversations can become a high-impact incident.
OpenClaw's own suggested fix is:
“Set each listed group/DM policy to "allowlist" and keep elevated allowlists extremely tight.”
Why does OpenClaw rate it critical?
OpenClaw's elevated mode docs say elevated mode "lets the agent break out of ordinary agent-configured sandboxing and run commands outside the sandbox instead". An open group or DM means anyone in it can send the agent instructions.
OpenClaw's audit triage order puts this combination first:
“Anything "open" + tools enabled: lock down DMs/groups first (pairing/allowlists), then tighten tool policy/sandboxing.”
If you need elevated mode
Keep it, but close the open policies first. Then keep the elevated allowlist short. The elevated mode docs list the gates that must all pass:
tools.elevated.enabledmust betrue.- The sender must be on
tools.elevated.allowFromfor that channel, for examplediscord: ["user-id-123"]. - Per-agent settings under
agents.entries.*.tools.elevated"can only further restrict".
The same docs say no bundled channel supplies a fallback allowlist today, so "every provider needs an explicit tools.elevated.allowFrom.<provider> entry".
Related findings
The same open policy often shows up in other findings too. OpenClaw's checks table links each of these to DM or group policy paths:
security.exposure.open_groups_with_runtime_or_fssecurity.exposure.open_channels_with_execsecurity.exposure.open_groups_with_control_plane_toolschannels.<provider>.dm.open, for an open DM policy
For the rest, see every audit finding, explained.
What if a stranger already messaged the agent?
The finding describes your config, not an event. If people you don't know could message the agent, follow OpenClaw's operator incident response plan in full. It covers containing access, rotating Gateway and provider credentials, reviewing logs, transcripts and recent config changes, and re-running openclaw security audit --deep.
Could it be a false alarm?
First, confirm what the config says:
openclaw config get channels openclaw config get tools.elevated
If a policy really is "open" and tools.elevated.enabled isn't false, the finding matches your config.
There's one open report about this checkId. Issue #175215, filed on October 17, 2026, describes an open channel bound only to a messaging-only agent that has elevated mode turned off for that agent. The reporter says the audit still lists other, more capable agents as reachable and raises this finding. OpenClaw's maintainers haven't confirmed it or shipped a fix.
If your setup matches that report, our guide to audit false alarms shows how to check a finding and suppress it properly. Don't suppress a critical finding you haven't confirmed.
What our free check covers
The free check at /check is paste-only. We check the text you paste: your openclaw security audit output. The check never connects to your Gateway or your chat channels, so it can't see who has messaged your agent. Nothing is installed into the agent you're checking. Green means no issues found in the checks we ran.
The free check puts the most important finding first, in plain English. The full report explains every finding and lists the fixes in order. The check doesn't detect malicious skills. We don't review the code inside skills or MCP servers ourselves, beyond what OpenClaw's own audit reports. The check is free, and a full report is $29. See Limits of the check.
Frequently asked questions
What does security.exposure.open_groups_with_elevated mean?
OpenClaw's audit found a channel where anyone can trigger the agent (a group or DM policy set to open) while elevated mode is available. OpenClaw says that, in those conversations, "a prompt injection … can become a high-impact incident". It rates the finding critical.
How do I fix security.exposure.open_groups_with_elevated?
Stop the Gateway. Set each policy the finding lists to a closed value: allowlist for groups, and pairing or allowlist for DMs. If you don't use elevated mode, also set tools.elevated.enabled to false. Validate the config, start the Gateway and run openclaw security audit --deep again.
I never turned on elevated mode. Why does the audit say it's enabled?
In OpenClaw 2026.10.5's source, this check counts elevated mode as enabled unless tools.elevated.enabled is set to false. So you can see the finding without having set elevated mode up. If you don't use it, setting tools.elevated.enabled to false clears that part of the finding.
Does this finding mean a stranger used my agent?
No. It describes your config, not an event. If people you don't know could message the agent, follow OpenClaw's operator incident response plan in full. It covers closing access, rotating Gateway and provider credentials, reviewing logs, transcripts and config changes, and re-running the audit.
Can your free check tell me who has messaged my agent?
No. We check the text you paste, which is where this finding comes from. The check never connects to your Gateway or your chat channels, so it can't see who has messaged your agent.
Sources
- OpenClaw docs: security audit checks (checkId catalog)
- OpenClaw docs: running the security audit (triage order, --fix)
- OpenClaw docs: elevated mode
- OpenClaw docs: groups (groupPolicy and group allowlists)
- OpenClaw docs: pairing (DM access approval)
- OpenClaw docs: operator incident response
- OpenClaw docs: openclaw config (CLI)
- OpenClaw source v2026.10.5: src/security/audit-extra.sync.ts (where the check is defined)
- OpenClaw source v2026.10.5: src/security/fix.ts (what --fix changes)
- GitHub issue #175215 (open): exposure checks ignore exact channel-to-agent bindings
Get the plain-English agent security brief
Agents & AI Security, explained: a short, plain-English email when we have something useful. Free. Unsubscribe any time.
We'll only use your email to send the newsletter. The newsletter is sent by beehiiv, which records opens and link clicks. Privacy policy.