GHSA-943Q-MWMV-HHVH
Vulnerability from github – Published: 2026-03-02 23:32 – Updated: 2026-03-02 23:32
VLAI
Summary
OpenClaw: Gateway /tools/invoke tool escalation + ACP permission auto-approval
Details
Summary
OpenClaw Gateway exposes an authenticated HTTP endpoint (POST /tools/invoke) intended for invoking a constrained set of tools. Two issues could combine to significantly increase blast radius in misconfigured or exposed deployments:
- The HTTP gateway layer did not deny high-risk session orchestration tools by default, allowing a caller with Gateway auth to invoke tools like
sessions_spawn/sessions_sendand pivot into creating or controlling agent sessions. - ACP clients could auto-approve permission requests for risky tools with insufficient user interaction/guardrails, reducing the friction that should normally prevent silent execution or mutation.
Impact
If the Gateway is reachable by an attacker and they obtain a valid Gateway token, they may be able to:
- Escalate from single-tool invocation to spawning/controlling sessions and reach command execution capabilities depending on tool policy and runtime environment.
- Perform cross-session message injection via
sessions_send. - In ACP-integrated scenarios, obtain unintended approvals for non-read/search tool permissions.
CVSS
CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H(8.8)
Affected versions
openclaw < 2026.2.14
Fixed in
openclaw >= 2026.2.14
Remediation
The default behavior is now hardened:
- PR #15390: deny high-risk tools over HTTP
/tools/invokeby default (withgateway.tools.{allow,deny}overrides) and harden ACP permission handling. - Commit
bb1c3dfe1: ACP clients now prompt for any non-read/search permission request (fail closed for mutating/execution/fetch operations). - Commit
539689a2f: security audit warns whengateway.tools.allowre-enables default-denied HTTP tools, since this can increase RCE blast radius if the Gateway is reachable. - Commit
153a7644e: ACP safe-kind inference is stricter to avoid accidental auto-approval due to substring matches (still auto-approves only confidentread/search).
Mitigations / deployment guidance
- Keep the Gateway loopback-only unless you have a strong reason not to:
gateway.bind="loopback"/openclaw gateway run --bind loopback. - Avoid exposing the Gateway directly to the public internet. Use an SSH tunnel or Tailscale to access a loopback-bound Gateway.
- Treat opting in to default-denied HTTP tools (via
gateway.tools.allow) as high-risk and audit such configurations carefully.
Credits
OpenClaw thanks @aether-ai-agent for reporting this issue and contributing remediation work.
Severity
8.8 (High)
{
"affected": [
{
"package": {
"ecosystem": "npm",
"name": "openclaw"
},
"ranges": [
{
"events": [
{
"introduced": "0"
},
{
"fixed": "2026.2.14"
}
],
"type": "ECOSYSTEM"
}
]
}
],
"aliases": [],
"database_specific": {
"cwe_ids": [
"CWE-78"
],
"github_reviewed": true,
"github_reviewed_at": "2026-03-02T23:32:22Z",
"nvd_published_at": null,
"severity": "HIGH"
},
"details": "## Summary\n\nOpenClaw Gateway exposes an authenticated HTTP endpoint (`POST /tools/invoke`) intended for invoking a constrained set of tools. Two issues could combine to significantly increase blast radius in misconfigured or exposed deployments:\n\n- The HTTP gateway layer did not deny high-risk session orchestration tools by default, allowing a caller with Gateway auth to invoke tools like `sessions_spawn` / `sessions_send` and pivot into creating or controlling agent sessions.\n- ACP clients could auto-approve permission requests for risky tools with insufficient user interaction/guardrails, reducing the friction that should normally prevent silent execution or mutation.\n\n## Impact\n\nIf the Gateway is reachable by an attacker and they obtain a valid Gateway token, they may be able to:\n\n- Escalate from single-tool invocation to spawning/controlling sessions and reach command execution capabilities depending on tool policy and runtime environment.\n- Perform cross-session message injection via `sessions_send`.\n- In ACP-integrated scenarios, obtain unintended approvals for non-read/search tool permissions.\n\n## CVSS\n\n- `CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H` (8.8)\n\n## Affected versions\n\n- `openclaw \u003c 2026.2.14`\n\n## Fixed in\n\n- `openclaw \u003e= 2026.2.14`\n\n## Remediation\n\nThe default behavior is now hardened:\n\n- PR #15390: deny high-risk tools over HTTP `/tools/invoke` by default (with `gateway.tools.{allow,deny}` overrides) and harden ACP permission handling.\n- Commit `bb1c3dfe1`: ACP clients now prompt for any non-read/search permission request (fail closed for mutating/execution/fetch operations).\n- Commit `539689a2f`: security audit warns when `gateway.tools.allow` re-enables default-denied HTTP tools, since this can increase RCE blast radius if the Gateway is reachable.\n- Commit `153a7644e`: ACP safe-kind inference is stricter to avoid accidental auto-approval due to substring matches (still auto-approves only confident `read/search`).\n\n## Mitigations / deployment guidance\n\n- Keep the Gateway loopback-only unless you have a strong reason not to: `gateway.bind=\"loopback\"` / `openclaw gateway run --bind loopback`.\n- Avoid exposing the Gateway directly to the public internet. Use an SSH tunnel or Tailscale to access a loopback-bound Gateway.\n- Treat opting in to default-denied HTTP tools (via `gateway.tools.allow`) as high-risk and audit such configurations carefully.\n\n## Credits\n\nOpenClaw thanks @aether-ai-agent for reporting this issue and contributing remediation work.",
"id": "GHSA-943q-mwmv-hhvh",
"modified": "2026-03-02T23:32:22Z",
"published": "2026-03-02T23:32:22Z",
"references": [
{
"type": "WEB",
"url": "https://github.com/openclaw/openclaw/security/advisories/GHSA-943q-mwmv-hhvh"
},
{
"type": "WEB",
"url": "https://github.com/openclaw/openclaw/pull/15390"
},
{
"type": "WEB",
"url": "https://github.com/openclaw/openclaw/commit/153a7644e"
},
{
"type": "WEB",
"url": "https://github.com/openclaw/openclaw/commit/539689a2f"
},
{
"type": "WEB",
"url": "https://github.com/openclaw/openclaw/commit/bb1c3dfe1"
},
{
"type": "WEB",
"url": "https://github.com/openclaw/openclaw/commit/ee31cd47b49f4b2f128a69a2a3745ca9db68b3be"
},
{
"type": "PACKAGE",
"url": "https://github.com/openclaw/openclaw"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H",
"type": "CVSS_V3"
}
],
"summary": "OpenClaw: Gateway /tools/invoke tool escalation + ACP permission auto-approval"
}
Loading…
Loading…
Experimental. This forecast is provided for visualization only and may change without notice. Do not use it for operational decisions.
Forecast uses a logistic model when the trend is rising, or an exponential decay model when the trend is falling. Fitted via linearized least squares.
Sightings
| Author | Source | Type | Date | Other |
|---|
Nomenclature
- Seen: The vulnerability was mentioned, discussed, or observed by the user.
- Confirmed: The vulnerability has been validated from an analyst's perspective.
- Published Proof of Concept: A public proof of concept is available for this vulnerability.
- Exploited: The vulnerability was observed as exploited by the user who reported the sighting.
- Patched: The vulnerability was observed as successfully patched by the user who reported the sighting.
- Not exploited: The vulnerability was not observed as exploited by the user who reported the sighting.
- Not confirmed: The user expressed doubt about the validity of the vulnerability.
- Not patched: The vulnerability was not observed as successfully patched by the user who reported the sighting.
Loading…
Loading…