Raw vector
CVSS:4.0/AV:N/AC:H/AT:P/PR:N/UI:N/VC:H/VI:H/VA:H/SC:N/SI:N/SA:N/E:U/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-2026-22708 is a high-severity External Control of System or Configuration Setting (CWE-15) vulnerability in Anysphere Cursor. Its CVSS base score is 7.2 (High).
Operationally, exploitation aligns with the MITRE ATT&CK technique Command and Scripting Interpreter (T1059); ranked at the 42th 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 LLM/Generative AI Risks risk domain.
The strongest mitigations our analysis identified map to AC-2 (Account Management) and AC-24 (Access Control Decisions) — 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-2026-22708 affects Cursor, an AI-powered code editor for programming, in versions prior to 2.3. The vulnerability resides in the Cursor Agent when operating in Auto-Run Mode with Allowlist mode enabled. In this configuration, certain shell built-ins can be executed without inclusion in the allowlist or requiring user approval, enabling attackers to leverage indirect or direct prompt injection to poison the shell environment. This involves setting, modifying, or removing environment variables that influence the behavior of trusted commands. The issue carries a CVSS v3.1 base score of 9.8 (AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H) and is associated with CWEs including CWE-15 (External Control of System or Configuration Setting), CWE-74 (Improper Neutralization of Special Elements), CWE-77/78 (Command Injection variants), CWE-94 (Code Injection), and CWE-269 (Privilege Context Switching).
Remote attackers require no privileges or user interaction to exploit this vulnerability over the network with low complexity. By crafting malicious prompts, they can manipulate environment variables, potentially altering the execution of trusted shell commands and leading to arbitrary code execution, data exfiltration, or system compromise within the poisoned environment.
The vulnerability is fully addressed in Cursor version 2.3, as detailed in the official security advisory at https://github.com/cursor/cursor/security/advisories/GHSA-82wg-qcm4-fp2w. Security practitioners should ensure users upgrade to 2.3 or later and review configurations for Auto-Run Mode and Allowlist usage, particularly in AI-assisted development workflows prone to prompt injection risks.
OWASP Top 10 for Web (2025)
EU & UK References
- 🇪🇺 ENISA EUVD: EUVD-2026-2678
Vulnerability Data
Cursor is a code editor built for programming with AI. Prior to 2.3, hen the Cursor Agent is running in Auto-Run Mode with Allowlist mode enabled, certain shell built-ins can still be executed without appearing in the allowlist and without…
more
requiring user approval. This allows an attacker via indirect or direct prompt injection to poison the shell environment by setting, modifying, or removing environment variables that influence trusted commands. This vulnerability is fixed in 2.3.
- CWE(s)
AI Security AnalysisAI
- AI Category
- Enterprise AI Assistants
- Risk Domain
- LLM/Generative AI Risks
- OWASP Top 10 for LLMs 2025
- None mapped
- Classification Reason
- Matched keywords: ai, prompt injection
Related Threats
MITRE ATT&CK Enterprise Techniques
CVEs Like This One
Affected Assets
Mitigating Controls
Control response
—
—
- 9 hardening rules · 7 OS baselines
V1.2.1V1.2.3V1.2.5V1.2.8
Mitigating Controls (NIST 800-53 r5) AI
Enforces approved authorizations for logical access to configuration resources so external parties cannot alter them.
Enforces documented access restrictions on all changes to system configuration, directly blocking unauthorized external control of settings.
Account management directly governs assignment and tracking of privileges so proper implementation stops the weakness from being introduced.
Requiring explicit access-control decisions ensures privileges are evaluated rather than assumed or omitted.
Separation of duties constrains how privileges may be assigned, reducing the chance of overly broad actor control.
Limits privileges to the minimum needed, reducing the set of users who can externally modify configuration.
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.
PR.AA-05 enforces least privilege/SoD and periodic reviews that directly remove most privilege-assignment defects, yet CWE-269 also covers escalation paths and role design outside a single access-management control.
Hardened configuration baselines and change controls directly limit external manipulation of settings.
Secure SDLC practices directly require input validation and output encoding that prevent injection flaws.
Runtime monitoring of software and data can detect anomalous command execution resulting from injection.
Identifying recorded vulnerabilities enables remediation of command-injection flaws before exploitation.
Formal change and exception management catches externally driven configuration alterations.
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.
Configuration management directly prevents external tampering with system settings.
Access rights assignment determines who may change system settings.
Privileged access rights reduce the number of users who can alter configuration.
Secure coding standards require proper escaping and parameterization of commands, directly eliminating CWE-77.
Security testing in development catches injection vulnerabilities before release.
Information access restriction limits who can view or modify configuration data.
Hardening callouts derived
Configuration rules from DISA STIG baselines that bear on weaknesses of the type cited by this CVE. Each rule is shown with the relationship its mapping actually records, against the CWE it was authored against. Derived via CVE→CWE over `controls_xwalks` (authoritative rows only; rows rated `none` are excluded).
Windows 10 (1 rule)
- V-220712 Only accounts responsible for the administration of a system must have Administrator rights on the system. prevents CWE-269
Windows 11 (1 rule)
- V-253269 Only accounts responsible for the administration of a system must have Administrator rights on the system. prevents CWE-269
Windows Server 2016 (1 rule)
- V-225007 Only administrators responsible for the member server or standalone or nondomain-joined system must have Administrator rights on the system. prevents CWE-269
Windows Server 2019 (1 rule)
- V-205746 Windows Server 2019 must only allow Administrators responsible for the member server or standalone or nondomain-joined system to have Administrator rights on the system. prevents CWE-269
Windows Server 2022 (1 rule)
- V-254428 Windows Server 2022 must only allow administrators responsible for the member server or standalone or nondomain-joined system to have Administrator rights on the system. prevents CWE-269