CVE-2026-33068
Anthropic Claude Code ≤ 2.1.53
Raw vector
CVSS:4.0/AV:N/AC:L/AT:P/PR:N/UI:P/VC:H/VI:H/VA:H/SC:N/SI:N/SA:N/E:X/CR:X/IR:X/AR:X/MAV:X/MAC:X/MAT:X/MPR:X/MUI:X/MVC:X/MVI:X/MVA:X/MSC:X/MSI:X/MSA:X/S:X/AU:X/R:X/V:X/RE:X/U:XSummary
CVE-2026-33068 is a high-severity Reliance on Untrusted Inputs in a Security Decision (CWE-807) vulnerability in Anthropic Claude Code. Its CVSS base score is 7.7 (High).
Operationally, exploitation aligns with the MITRE ATT&CK technique Exploit Public-Facing Application (T1190); ranked at the 27th percentile by exploit likelihood (below the median); it is not currently listed in the CISA KEV catalog.
This vulnerability is AI-related — categorised as Enterprise AI Assistants; in the Supply Chain and Deployment risk domain.
The strongest mitigations our analysis identified map to AC-24 (Access Control Decisions) and SI-10 (Information Input Validation) — see the control section below for these in your framework.
Deeper analysis AI-assisted summary
Synthesised by an AI model from the NVD description and linked references — a reading aid, not an authoritative source.
CVE-2026-33068 is a vulnerability in Claude Code, an agentic coding tool from Anthropic, affecting versions prior to 2.1.53. The issue arises because the tool resolves the permission mode from settings files, including the repository-controlled .claude/settings.json, before determining whether to display the workspace trust confirmation dialog. A malicious repository can commit a .claude/settings.json file setting permissions.defaultMode to bypassPermissions, which causes the trust dialog to be silently skipped upon first opening the repository. This flaw is classified under CWE-807 (Reliance on Untrusted Inputs in a Security Decision) and carries a CVSS v3.1 base score of 8.8 (AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H).
Attackers who control a repository can exploit this vulnerability by embedding the malicious settings file in their code. Victims, such as developers who clone or open the attacker-controlled repository in Claude Code, would unknowingly enter permissive mode without the trust confirmation prompt. This enables the repository to gain tool execution privileges without explicit user consent, potentially allowing arbitrary code execution or other malicious actions within the tool's environment.
The vulnerability has been patched in Claude Code version 2.1.53. Additional details on the fix and affected versions are available in the GitHub security advisory at https://github.com/anthropics/claude-code/security/advisories/GHSA-mmgp-wc2j-qcv7.
OWASP Top 10 for Web (2025)
EU & UK References
- 🇪🇺 ENISA EUVD: EUVD-2026-13631
Vulnerability Data
Claude Code is an agentic coding tool. Versions prior to 2.1.53 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…
more
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. This issue has been patched in version 2.1.53.
- CWE(s)
AI Security AnalysisAI
- AI Category
- Enterprise AI Assistants
- Risk Domain
- Supply Chain and Deployment
- OWASP Top 10 for LLMs 2025
- None mapped
- Classification Reason
- Matched keywords: claude
Related Threats
MITRE ATT&CK Enterprise Techniques
CVEs Like This One
Affected Assets
Mitigating Controls
Control response
—
—
- 1 hardening rule · 1 OS baseline
—
Mitigating Controls (NIST 800-53 r5) AI
Explicitly requires that every access decision be based on authoritative, protected decision data rather than caller-supplied inputs.
Input validation can reject malformed or attacker-controlled values before they reach a security decision point.
Access enforcement requires decisions to be made from trusted policy data rather than modifiable client-supplied inputs.
Information-flow enforcement applies rules to validated, internal attributes instead of untrusted external values.
Mitigating Controls (NIST CSF 2.0) AI
Derived directly from the weakness types (CWEs) cited in the NVD entry via our AI-authored CWE→CSF cross-walk (authority under review) — links open the control.
Strong authentication mechanisms directly prevent security decisions from depending on modifiable, untrusted inputs.
Verification and protection of identity assertions stops reliance on attacker-controlled values for authorization decisions.
Proper management of identities/credentials reduces the chance that security decisions will be driven by untrusted inputs.
Policy-driven, least-privilege authorization reduces opportunities to bypass controls via tampered inputs.
Mitigating Controls (ISO/IEC 27001:2022 Annex A) AI
Derived directly from the weakness types (CWEs) cited in the NVD entry via our AI-authored CWE→ISO cross-walk (authority under review) — links open the control.
Secure coding standards directly prohibit using untrusted inputs for security-critical decisions.
Security testing in development catches input-validation flaws before deployment.
Monitoring can detect exploitation but does not stop the underlying weakness.
Application security requirements explicitly call for validation of inputs used in security decisions.
Secure architecture principles mandate treating all external inputs as untrusted.
Access-control rules can be bypassed if they rely on untrusted inputs.