CVE-2026-12823: Browserbase Skills Permissions Misconfiguration (Local Information Disclosure)

Technical summary

CVE-2026-12823 stems from improper permission defaults in the Autobrowse Trace Artifact Handler within Browserbase Skills. The vulnerability is classified under CWE-266 (Incorrect Privilege Assignment) and CWE-276 (Incorrect Default Permissions), indicating that sensitive artifacts or data structures are accessible to local users with insufficient privilege separation. An authenticated local attacker can read confidential data without requiring user interaction. The CVSS 3.1 score of 3.3 (LOW) reflects the local-only attack vector and limited confidentiality impact, with no availability or integrity compromise.

Business impact

For organizations deploying Browserbase Skills in shared or multi-tenant environments, this vulnerability could expose sensitive automation traces and debugging artifacts to unauthorized local users. This may include API credentials, session tokens, or internal browsing logic embedded in trace logs. While not a critical risk to availability or system integrity, the information disclosure could support reconnaissance for further attacks or violate data residency and confidentiality policies. The business impact is primarily to data sensitivity rather than operational disruption.

Affected systems

Browserbase Skills versions up to and including 20260526 are affected. The vulnerability is specific to the Browserbase/skills project (not Browserbase itself, despite initial confusion in the CVE record). Organizations running this component should verify their installed version against 20260526. Verify against the vendor advisory for specific patched version availability and any version numbering schemes used by the project.

Exploitability

Exploitation requires local system access and valid user credentials; it cannot be triggered remotely or without authentication. Public proof-of-concept code is available, reducing the barrier to exploitation for insiders or attackers with local foothold. However, the low severity and local-only nature mean this is not a vector for widespread compromise. Exploitation is opportunistic rather than mass-exploitable.

Remediation

Upgrade Browserbase Skills to a version later than 20260526. Verify the exact patched version against the vendor's official advisory. Until patching is possible, restrict local system access to trusted users only, segment deployment environments by sensitivity level, and monitor local user activity for unauthorized access to artifact directories or trace files.

Patch guidance

Apply the latest version of Browserbase Skills beyond 20260526 as confirmed by the vendor's official release notes. Test the upgrade in a development environment first to ensure compatibility with your automation workflows. The vendor was contacted early during disclosure, so a patch should be available or imminently released. Verify patch availability and version numbers in the official Browserbase Skills repository or advisory documentation.

Detection guidance

Monitor for unusual local file access patterns targeting Autobrowse Trace Artifact Handler directories. Look for unprivileged local users reading trace files, artifact logs, or debugging output they should not have access to. File integrity monitoring on artifact storage locations can detect unauthorized access attempts. Review system access logs for privilege escalation attempts or lateral movement from shared systems following artifact access.

Why prioritize this

While the CVSS score is low, prioritize this based on your environment's sensitivity to insider threats and multi-tenant deployment patterns. If Browserbase Skills runs in isolated, single-user environments with strong access controls, remediation is routine maintenance. If deployed in shared infrastructure or accessed by less-trusted local users, move it up in your patch queue to prevent confidentiality breaches of sensitive automation data.

Risk score, explained

CVSS 3.1 score of 3.3 (LOW) reflects: local-only attack vector (AV:L), low complexity with no special conditions (AC:L), requirement for low-level privileges and authentication (PR:L), no user interaction needed (UI:N), no system boundary crossing (S:U), and confidentiality impact only (C:L, I:N, A:N). The score appropriately penalizes the vulnerability for its local scope while acknowledging the information disclosure risk.

Frequently asked questions