/

CVE-2026-35653

Severity:

HIGH RISK

(8.1)

Incorrect Authorization in OpenClaw /reset-profile Endpoint

CVE-2026-35653

Analyzed

Vulnerability type

Category:

Incorrect Authorization

CWE:

CWE-863

Impact:

Privilege Escalation and Unauthorized Profile Manipulation

Severity:

HIGH RISK

(8.1)

CVSS version:

3.1

Catch this CVE while code is being written

Detect vulnerable code directly in pull requests, so CVEs are fixed before merge.

Summary

OpenClaw versions before 2026.3.24 contain an incorrect authorization flaw in the POST /reset-profile endpoint that allows certain authenticated callers to bypass intended profile mutation restrictions.

Attackers with operator.write access to the browser.request surface can invoke /reset-profile to stop the running browser, close Playwright connections, and move profile directories to Trash, crossing designed privilege boundaries.

Affected context

This vulnerability appears in OpenClaw deployments where:

  • The POST /reset-profile endpoint is exposed internally or via an API surface such as browser.request
  • Roles or capabilities with operator.write on browser.request exist (for automation, operational tooling, or integrations)
  • The authorization layer treats/operator.write as sufficient to operate on browser-related actions but is expected to restrict destructive profile operations

In these conditions, a user or integration with limited operational privileges can trigger profile resets that should be reserved for higher-privileged maintenance or administrative paths.

Exploit preconditions

  • Attack surface: The OpenClaw API, specifically the browser.request surface that can reach POST /reset-profile
  • Required access: An authenticated principal (user, service account, or integration) with operator.write permissions on browser.request
  • Trigger point: Issuing a POST request to /reset-profile through the browser.request mechanism
  • Why exploitation is possible: The authorization logic for /reset-profile does not distinguish between routine operator.write actions and higher-risk profile-destructive operations, allowing callers to perform actions beyond their intended privilege level.

Why this happens

The underlying bug is an authorization policy mismatch between what a given role (operator.write) is supposed to do and what the /reset-profile endpoint actually allows.

Mapped to CWE-863: Incorrect Authorization, the system validates that a caller is authenticated and holds some level of permission, but fails to enforce the correct, more restrictive privilege required for disruptive profile resets.

Instead of performing fine-grained checks on who may stop browsers, terminate Playwright connections, and move profile directories, the endpoint trusts a broader capability on the browser.request surface, effectively expanding the caller's power beyond its intended scope.

Potential impact

  • An operator-level account may stop running browsers unexpectedly, disrupting automated tests, scraping jobs, or user sessions.
  • Attackers could close active Playwright connections, interrupting CI pipelines or browser-based workflows.
  • Profiles and their directories could be moved to Trash, leading to data loss or forced recreation of automation environments.
  • Privilege boundaries between operational and administrative roles could be eroded, enabling broader system misuse.
  • Repeated exploitation could be used as a denial-of-service technique against browser-based infrastructure managed by OpenClaw.

General mitigation guidance

Mitigation generally focuses on tightening authorization around sensitive endpoints and clarifying privilege boundaries:

  • Define a distinct, higher-privilege capability or role specifically for destructive or reset-style operations such as /reset-profile, separate from routine operator.write permissions.
  • Implement fine-grained authorization checks per endpoint, basing access on least privilege rather than broad, surface-level permissions (e.g., browser.request).
  • Introduce explicit policy rules that classify actions like stopping browsers, terminating connections, and manipulating profile directories as privileged maintenance operations.
  • Log and monitor all invocations of high-impact endpoints, and consider adding interactive confirmation or multi-party approval workflows where appropriate.
  • Where feasible, separate read/observe, routine operational actions, and destructive maintenance operations into distinct permission sets to reduce blast radius.

How teams detect issues like this

Detection can rely on both code-centric and operational approaches:

  • Static analysis and code review:
  • Flag endpoints that perform destructive operations (reset, delete, move-to-trash, stop, terminate) and ensure they reference the narrowest appropriate permission.
  • Look for shared authorization checks reused across both low-risk and high-risk endpoints, indicating potential over-broad privileges.
  • Review controller or handler code for any mismatch between documented role semantics and actual enforced checks.
  • Policy and configuration review:
  • Inspect role and permission definitions to confirm that operator-like roles cannot invoke profile-destructive operations.
  • Validate that API gateway or middleware policies distinguish between generic browser actions and sensitive lifecycle/profile operations.
  • Runtime monitoring:
  • Instrument logging for all calls to /reset-profile, including caller identity, source, and context.
  • Set alerts on unexpected patterns, such as frequent profile resets, resets from unusual service accounts, or resets during business-critical windows.
  • Automated security review tools can help identify authorization hotspots by scanning for endpoints tied to high-impact actions without correspondingly strict access control checks.

Sources

SourceLink
github.comgithub.com/openclaw/openclaw/commit/4dcc39c25c6cc63fedfd004f52d173716576fcf0
github.comgithub.com/openclaw/openclaw/commit/e7d11f6c33e223a0dd8a21cfe01076bd76cef87a
github.comgithub.com/openclaw/openclaw/security/advisories/GHSA-xp9r-prpg-373r
vulncheck.comvulncheck.com/advisories/openclaw-incorrect-authorization-in-post-reset-profile-via-browser-request

Frequently asked

questions

It can be, depending on how OpenClaw is deployed. If the browser.request surface is exposed over a network and accessible to external or semi-trusted clients, any party with operator.write on that surface may remotely trigger /reset-profile. In more locked-down setups, exploitation is typically limited to internal or service accounts.

Exploitation requires a valid authenticated identity (user or service account) that has operator.write permissions on the browser.request surface. The vulnerability does not permit completely unauthenticated access; it lets authenticated but under-privileged identities perform actions that should require stronger permissions.

The most robust fix is to implement stricter, explicit authorization on sensitive endpoints like /reset-profile. This usually means introducing a dedicated, higher-privilege role or capability (e.g., admin.maintenance or browser.profile.reset) and requiring that permission instead of relying on broader operator.write access. Aligning endpoint-level checks with clearly documented role semantics reduces the risk of similar issues.

Prevention hinges on designing access control with least privilege and clear privilege tiers. Practical measures include:

  • Classifying endpoints by impact (read-only, routine operational, destructive/maintenance) and assigning distinct permissions to each class.
  • Using centralized authorization middleware with policies that can be reviewed and tested, rather than ad hoc per-endpoint checks.
  • Including access-control reviews in code review checklists and security design reviews.
  • Employing automated policy-as-code and static analysis where possible, to surface endpoints performing high-impact operations without appropriately restrictive authorization.
Find out what's already
exploitable in your codebase.