What is the CVSS score of CVE-2026-33068?

CVE-2026-33068 has a CVSS 3.1 base score of 8.8 (High). CVSS vector: CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H. EPSS exploitation probability: 0.1%.

Is there a public PoC exploit available for CVE-2026-33068?

Yes, a public proof-of-concept exploit is available for CVE-2026-33068. Prioritise patching immediately. EPSS: 0.1%.

How to fix or mitigate CVE-2026-33068?

Within 24 hours: Inventory all internal use of @anthropic-ai/claude-code npm package and identify affected development teams. Within 7 days: Issue security advisory to developers prohibiting Claude Code usage on untrusted/external repositories and enforce code review policies for repository sources. Within 30 days: Monitor Anthropic security advisories for patch availability; evaluate alternative coding assistants with stronger trust controls; implement network segmentation to limit development environment lateral movement potential.

CVE-2026-33068

HIGH

Reliance on Untrusted Inputs in a Security Decision (CWE-807)

2026-03-19 https://github.com/anthropics/claude-code

GHSA-mmgp-wc2j-qcv7

Authentication Bypass

8.8

CVSS 3.1 · GitHub Advisory

Severity by source

GitHub Advisory PRIMARY

8.8 HIGH

AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H

Primary rating from GitHub Advisory · only source for this CVE.

CVSS VectorGitHub Advisory

Attack Vector

Network

Attack Complexity

Low

Privileges Required

None

User Interaction

Required

Scope

Unchanged

Confidentiality

High

Integrity

High

Availability

High

Lifecycle Timeline

Patch released

Mar 31, 2026 - 21:13 nvd

Patch available

Analysis Generated

Mar 19, 2026 - 12:45 vuln.today

CVE Published

Mar 19, 2026 - 12:42 nvd

HIGH 8.8

DescriptionGitHub Advisory

Claude Code resolved the permission mode from settings files, including the repo-controlled .claude/settings.json, before determining whether to display the workspace trust confirmation dialog. A malicious repository could set permissions.defaultMode to bypassPermissions in its committed .claude/settings.json, causing the trust dialog to be silently skipped on first open. This allowed a user to be placed into a permissive mode without seeing the trust confirmation prompt, making it easier for an attacker-controlled repository to gain tool execution without explicit user consent.

Users on standard Claude Code auto-update have received this fix already. Users performing manual updates are advised to update to the latest version.

Thank you to hackerone.com/cantina_xyz for reporting this issue.

AnalysisAI

Claude Code, an AI coding assistant, contains an authentication bypass vulnerability where malicious repositories can silently skip the workspace trust confirmation dialog by setting permissions.defaultMode to bypassPermissions in a committed .claude/settings.json file. This affects users of the @anthropic-ai/claude-code npm package who open untrusted repositories. An attacker can place users into a permissive execution mode without explicit consent, enabling tool execution without the user seeing trust prompts, though no evidence of active exploitation or public proof-of-concept is currently available.

Technical ContextAI

This vulnerability affects the Claude Code extension (pkg:npm/@anthropic-ai_claude-code), an AI-powered coding tool. The root cause is classified as CWE-807 (Reliance on Untrusted Inputs in a Security Decision), where the application incorrectly trusts repository-controlled configuration files when making security decisions. Specifically, Claude Code reads permission settings from .claude/settings.json files within repositories before validating whether to display workspace trust dialogs. By resolving the permission mode from attacker-controlled settings files, the application allows repository content to dictate security policy enforcement, violating the principle that security decisions should not rely on untrusted inputs.

RemediationAI

Users on Claude Code's standard auto-update channel have already received the security fix automatically and require no manual action. Users who have disabled automatic updates or perform manual installations should immediately update to the latest version of @anthropic-ai/claude-code through their package manager or extension marketplace. Refer to the official security advisory at https://github.com/anthropics/claude-code/security/advisories/GHSA-mmgp-wc2j-qcv7 for detailed update instructions. As a temporary mitigation until updating is possible, users should exercise extreme caution when opening repositories from untrusted sources, manually review any .claude/settings.json files in repositories before opening them in Claude Code, and ensure workspace trust prompts are carefully reviewed when they do appear.