Cursor
Anysphere · IDE · homepage
Minimum safe version is set by CVE-2026-73217 (Sandbox escape via tampered Python virtual environments, fixed in 3.1.2): the highest fix among Cursor's medium-or-higher advisories. Only advisories for Cursor itself are counted; other components and ecosystem projects are listed separately below. 3 medium-or-higher advisories list no fixed version: check the vendor's guidance before relying on the minimum.
Advisories · Data exposure · Hardening
Advisories
| ID | Title | Severity | Fixed in | Published |
|---|---|---|---|---|
| CVE-2026-73217 | Sandbox escape via tampered Python virtual environments · sets minimum safe version | High | 3.1.2 | 2026-07-14 |
| CVE-2026-73218 | Sandbox escape via launching privileged containers | High | — | 2026-07-14 |
| CVE-2026-61613 | Cloud Agent Browser Sandbox Escape | High | — | 2026-07-06 |
| CVE-2026-50548 | Cursor Desktop sandbox escape via agent-controlled working directory | Critical | 3.0 | 2026-06-05 |
| CVE-2026-50549 | Cursor Desktop sandbox escape via symlink and failed path canonicalization | Critical | 3.0 | 2026-06-05 |
| CVE-2026-48124 | Cursor Desktop sandbox escape via Claude hook configuration | High | 3.0.0 | 2026-05-21 |
| CVE-2026-31854 | Arbitrary Code Execution via Prompt Injection and Whitelist Bypass | High | 2.0 | 2026-03-09 |
| CVE-2026-26268 | Sandbox escape via Git hooks | High | 2.5 | 2026-02-13 |
| CVE-2026-22708 | Terminal Tool Allowlist Bypass via Environment Variables | High | 2.3 | 2026-01-14 |
| CVE-2025-64110 | Cursorignore Bypass via New Cursorignore Write | High | 2.0 | 2025-11-03 |
Show 16 older advisories
| ID | Title | Severity | Fixed in | Published |
|---|---|---|---|---|
| CVE-2025-64106 | Speedbump Modal Bypass in Cursor MCP Server Deep-Link | High | 2.0 | 2025-11-03 |
| CVE-2025-64108 | Sensitive File Modification - NTFS Path Quirks | High | 2.0 | 2025-11-03 |
| CVE-2025-64107 | Sensitive File Protection Bypass - Path Manipulation Using Backslashes on Windows | High | 2.0 | 2025-11-03 |
| CVE-2025-59944 | Cursor IDE - Sensitive File Overwrite Bypass | High | 1.7 | 2025-10-02 |
| CVE-2025-61590 | RCE via .code-workspace files using Prompt Injection | High | 1.7 | 2025-10-02 |
| CVE-2025-61589 | Potential Information Leakage via Mermaid Diagram | Medium | 1.7 | 2025-10-02 |
| CVE-2025-54135 | Arbitrary code execution from Cursor Agent through a prompt injection via MCP Special Files | High | 1.3.9 | 2025-08-02 |
| CVE-2025-54130 | Arbitrary code execution from Cursor Agent through a prompt injection via Editor Special Files | High | 1.3.9 | 2025-08-02 |
| CVE-2025-54131 | Bypassing allow list to execute an arbitrary command | Medium | 1.3 | 2025-08-01 |
| CVE-2025-54132 | Arbitrary Image Fetch in Mermaid Diagram Tool | Medium | 1.3 | 2025-08-01 |
| CVE-2025-54133 | MCP Install Deeplink Did Not Show Arguments on User-Dialog | Medium | 1.3 | 2025-08-01 |
| CVE-2025-54136 | Modification of MCP Server Definitions Bypasses Manual Re-approval | High | 1.3 | 2025-08-01 |
| CVE-2025-49150 | Potential Information Leakage using JSON schema in Cursor Agent | Medium | 0.51.0 | 2025-06-11 |
| CVE-2025-32018 | Arbitrary file write from Cursor Agent through a prompt injection from malicious @Docs | High | 0.48.7 | 2025-04-07 |
| CVE-2024-48919 | RCE via Prompt Injection Into Cursor's Terminal Cmd-K | High | — | 2024-10-22 |
| CVE-2024-45599 | TCC Bypass in Cursor's macOS Application | Low | 0.41.0 | 2024-09-24 |
Related products and ecosystem projects
Advisories for other Anysphere components or third-party projects that build on Cursor. Not counted above and not used for the minimum safe version.
| ID | Component | Title | Severity | Fixed in | Published |
|---|---|---|---|---|---|
| CVE-2025-64109 | Cursor CLI (Beta) | Command Injection via Untrusted MCP Configuration in Cursor CLI Beta | High | 2025.09.17-25b418f | 2025-11-03 |
| CVE-2025-61593 | Cursor CLI | Cursor CLI Agent - Sensitive File Overwrite Bypass | High | 2025.09.17-25b418f | 2025-10-02 |
| CVE-2025-61591 | Cursor CLI | Remote Code Execution in Cursor CLI via Cursor Agent MCP OAuth2 Communication | High | 2025.09.17-25b418f | 2025-10-02 |
| CVE-2025-61592 | Cursor CLI | Arbitrary code execution Permissive CLI Config in Cursor CLI | High | 2025.09.17-25b418f | 2025-10-02 |
| GHSA-c43p-6fv2-6gr2 | Cursor API | IDOR in Usage API Leading to Unauthorized Data Exposure | Medium | — | 2025-06-19 |
| GHSA-rjmc-526x-8653 | Cursor API | IDOR in Usage Events API Lets Users Access Teammates' Activity Data | Medium | — | 2025-06-19 |
| GHSA-g4ff-54cv-h6f9 | Cursor Billing | Any Team Member Can Access Billing Link | Low | — | 2024-11-27 |
Hardening checklist
- Keep auto-update enabled and verify the installed version against the minimum safe version.
- Do not open untrusted repositories with agent mode or auto-approve enabled.
- Restrict MCP servers to an allowlist; review each server's permissions and publisher.
- Never place secrets in project files the agent can read; use a secrets manager or env injection.
- Disable auto-run for terminal commands in untrusted workspaces; enable Workspace Trust.
Data exposure
What Cursor keeps on a developer machine, what it sends, and the settings that reduce exposure. Local transcripts can contain anything pasted into a session, including secrets: include these paths in secret scanning and offboarding.
Stored locally
| What | Where |
|---|---|
| Local SQLite state database. Researchers found it holds session tokens and third-party API keys without OS-level protection, readable by any installed extension (reported Feb 2026, unfixed as of Apr 2026). | macOS ~/Library/Application Support/Cursor/User/globalStorage/state.vscdb |
| User data, settings and workspace storage (Windows). | %APPDATA%\Cursor\ |
| Agent worktrees: full checkouts of your repository created by the CLI or editor. | ~/.cursor/worktrees/<reponame>/<name> |
| Cursor CLI global configuration. | ~/.cursor/cli-config.json |
| Instant Grep search index, built and queried on the machine. | Local (path not documented) |
Sent to the vendor
Prompts and code context go through Cursor's backend to model providers, even when you use your own API key. File contents are cached temporarily on Cursor servers, encrypted with client-generated keys. Cloud Agents store encrypted repository copies while they run. Cursor does not upload code or embeddings to build its search index.
Retention and training
With Privacy Mode on, Cursor does not train on your data and holds zero data retention agreements with providers, except data flagged by abuse classifiers. With Privacy Mode off, Cursor may store and train on prompts, code and editor actions. Privacy Mode is on by default for Enterprise; models that need provider retention require admin approval.
Settings that reduce exposure
- Privacy Mode
- No training on your data and ZDR with model providers for Cursor-routed models.
.cursorignore- Blocks listed files from Agent, Tab and @ references; terminal and MCP tools cannot enforce it.
- Global ignore list (user settings)
- Applies ignore patterns such as
**/.envto all projects; empty by default. - Run Mode: Auto-review with sandboxing
- Shell commands run sandboxed where possible instead of executing everything unattended.
- Account deletion (Dashboard)
- Removes the user's account and associated data within 30 days.
For administrators
- Enforce Privacy Mode team-wide so members cannot disable it.
AllowedTeamIdMDM policy blocks personal accounts (which may lack Privacy Mode) on corporate devices.AllowedExtensionsMDM policy allowlists extension publishers.- Restrict personal API keys (BYOK); with your own keys, retention follows your own provider agreement.
- Model access control, Git repository blocklist and MCP allowlist.
- Shared transcript retention (Enterprise, 30 to 365 days; off by default) and Cloud Agent retention windows.
- Audit logs streamed to SIEM; SSO and SCIM;
x-cursor-allowed-tenantsheader for tenancy restriction.
Sources: Cursor: Data Use & Privacy Overview · Cursor: Security · Cursor docs: Privacy and Data Governance · Cursor docs: Security and Privacy Hardening · Cursor docs: Ignore file · Cursor docs: Identity and Access Management (MDM policies) · Cursor docs: Endpoint Security Configuration · Cursor docs: Search (Instant Grep privacy) · Cursor docs: CLI configuration · Akamai (LayerX research): CursorJacking — API key theft by rogue extensions · reviewed 2026-11-09
Rolling these tools out? See the playbook Secure AI coding tools in your organisation.
Lock it down: AI hardening kit (managed settings and MCP allowlist generators, free, in your browser).
Check your version
Via the public API:
curl -s "https://threatdeck.net/api/v1/tools/cursor/check?version=<your-version>" Or ask your agent through the ThreatDeck MCP server (https://threatdeck.net/mcp):
check_ai_tool_version { "slug": "cursor", "version": "<your-version>" } Updated 2026-11-22