Claude Code
Anthropic · CLI agent · homepage · anthropics/claude-code
Minimum safe version is set by CVE-2026-88968 (TOCTOU Race in Edit Tool Allows File Writes Outside the Workspace, fixed in 2.1.251): the highest fix among Claude Code's medium-or-higher advisories. Only advisories for Claude Code itself are counted; other components and ecosystem projects are listed separately below.
Advisories · Data exposure · Hardening
Advisories
| ID | Title | Severity | Fixed in | Published |
|---|---|---|---|---|
| CVE-2026-88968 | TOCTOU Race in Edit Tool Allows File Writes Outside the Workspace · sets minimum safe version | High | 2.1.251 | 2026-10-26 |
| CVE-2026-86063 | Plugin Commit SHA Pin Bypass Allows Installation of Attacker-Controlled Code | High | 2.1.179 | 2026-10-09 |
| CVE-2026-103435 | Arbitrary File Write via Write-Time Symlink Following (TOCTOU) in Claude Code | High | 2.1.129 | 2026-10-05 |
| CVE-2026-103012 | Managed Settings Not Applied When a Stored API Key Was Preferred Over Enterprise Sign-In | Low | 2.1.260 | 2026-09-29 |
| CVE-2026-46406 | Insecure Temporary File in /copy Command Enables Response Disclosure and Symlink-Based File Write | Medium | 2.1.128 | 2026-06-25 |
| CVE-2026-55607 | Sandbox Escape via Git Worktree Path Confusion Allows Unsandboxed Code Execution | High | 2.1.163 | 2026-06-25 |
| CVE-2026-54316 | Out-of-Band Data Exfiltration via Pre-Approved HuggingFace Domain in WebFetch | Medium | 2.1.163 | 2026-06-13 |
| CVE-2026-40068 | Trust Dialog Bypass via Git Worktree Spoofing Allows Arbitrary Code Execution | High | 2.1.84 | 2026-04-24 |
| CVE-2026-39861 | Sandbox Escape via Symlink Following Allows Arbitrary File Write Outside Workspace | High | 2.1.64 | 2026-04-20 |
| CVE-2026-35603 | Insecure System-Wide Configuration Loading Enables Local Privilege Escalation on Windows | Medium | 2.1.75 | 2026-04-17 |
Show 22 older advisories
| ID | Title | Severity | Fixed in | Published |
|---|---|---|---|---|
| CVE-2026-33068 | Workspace Trust Dialog Bypass via Repo-Controlled Settings File | High | 2.1.53 | 2026-03-18 |
| CVE-2026-25722 | Command Injection via Directory Change Bypasses Write Protection | High | 2.0.57 | 2026-02-06 |
| CVE-2026-25723 | Command Injection via Piped sed Command Bypasses File Write Restrictions | High | 2.0.55 | 2026-02-06 |
| CVE-2026-25725 | Sandbox Escape via Persistent Configuration Injection in settings.json | High | 2.1.2 | 2026-02-06 |
| CVE-2026-25724 | Permission Deny Bypass Through Symbolic Links | Low | 2.1.7 | 2026-02-06 |
| CVE-2026-24887 | Command Injection in find Command Bypasses User Approval Prompt | High | 2.0.72 | 2026-02-03 |
| CVE-2026-24053 | Path Restriction Bypass via ZSH Clobber Allows Arbitrary File Writes | High | 2.0.74 | 2026-02-03 |
| CVE-2026-24052 | Domain Validation Bypass Allows Automatic Requests to Attacker-Controlled Domains | High | 1.0.111 | 2026-02-03 |
| CVE-2026-21852 | Malicious repo configuration can trigger data leakage via environment configuration used before trust confirmation | Medium | 2.0.65 | 2026-01-20 |
| CVE-2025-66032 | Command Validation Bypass Allows Arbitrary Code Execution | High | 1.0.93 | 2025-12-03 |
| CVE-2025-64755 | Sed Command Validation Bypass Allows Arbitrary File Writes | High | 2.0.31 | 2025-11-20 |
| CVE-2025-65099 | Command execution prior to Claude Code startup trust dialog | High | 1.0.39 | 2025-11-19 |
| CVE-2025-59536 | Command execution prior to Claude Code startup trust dialog | High | 1.0.111 | 2025-10-03 |
| CVE-2025-59829 | Permission deny bypass through symlink | Low | 1.0.120 | 2025-10-03 |
| CVE-2025-59828 | Claude Code Vulnerable to Arbitrary Code Execution via Plugin Autoloading with Specific Yarn Versions | High | 1.0.39 | 2025-09-24 |
| CVE-2025-58764 | Command Injection in Claude Code rg command allowed bypass of user approval prompt for command execution | High | 1.0.105 | 2025-09-09 |
| CVE-2025-59041 | Claude Code vulnerable to arbitrary code execution caused by maliciously configured git email | High | 1.0.105 | 2025-09-09 |
| GHSA-ph6w-f82w-28w6 | Claude Code Vulnerable to Arbitrary Code Execution Due to Insufficient Startup Warning | High | 1.0.87 | 2025-09-02 |
| CVE-2025-55284 | Permissive Default Allowlist Enables Unauthorized File Read and Network Exfiltration in Claude Code | High | 1.0.4 | 2025-08-15 |
| CVE-2025-54794 | Path Restriction Bypass in Claude Code Research Preview could allow unauthorized file access when path prefixes collide | High | 0.2.111 | 2025-08-01 |
| CVE-2025-54795 | Command Injection in Claude Code echo command allowed bypass of user approval prompt for command execution | High | 1.0.20 | 2025-08-01 |
| CVE-2025-52882 | Claude Code IDE extensions allow websocket connections from arbitrary origins | High | 1.0.24 | 2025-06-23 |
Related products and ecosystem projects
Advisories for other Anthropic components or third-party projects that build on Claude Code. Not counted above and not used for the minimum safe version.
| ID | Component | Title | Severity | Fixed in | Published |
|---|---|---|---|---|---|
| CVE-2026-77390 | pmo-platform | pmo-platform is a modular PMO and release-management platform for Claude Code. Prior to 3.73, release/skills/pmo-skill-refiner/eval-viewer/g… | Medium | — | 2026-11-20 |
| CVE-2026-77389 | pmo-platform | pmo-platform is a modular PMO and release-management platform for Claude Code. Prior to 3.73, core/hooks/block-fragile-refs.sh and core/hook… | Low | — | 2026-11-20 |
| CVE-2026-77350 | pmo-platform | pmo-platform is a modular PMO and release-management platform for Claude Code. Prior to 3.69.1, core/hooks/block-credential-reads.sh, core/h… | High | — | 2026-11-20 |
| CVE-2026-77349 | pmo-platform | pmo-platform is a modular PMO and release-management platform for Claude Code. From 3.48 until 3.69.1, release/skills/pmo-skill-refiner/eval… | Medium | — | 2026-11-20 |
| CVE-2026-75612 | yt-mcp | yt-mcp is a YouTrack MCP server for Claude Code. Prior to 1.18.1, add_attachment in src/yt_mcp/tools/attachments.py accepts a caller-control… | High | — | 2026-11-18 |
| CVE-2026-62331 | praktor | praktor is a multi-agent Claude Code orchestrator with Telegram input and output, Docker isolation, and a Mission Control UI. Prior to 0.9.1… | High | — | 2026-10-30 |
| CVE-2026-96620 | Claude Agent SDK | Claude Agent SDK is a Python SDK for Claude Code. Prior to 0.2.121, the resume option is passed to the Claude CLI as a separate token withou… | Critical | — | 2026-10-16 |
| CVE-2026-88066 | Claude Desktop | Uncontrolled Search Path Enables Arbitrary Code Execution via Malicious Git Executable | High | 1.8089.0 | 2026-10-13 |
| CVE-2026-100585 | OpenClaw | OpenClaw (npm package `openclaw`) before 2026.7.1 fails to enforce the owner-only authorization requirement for Claude Code permission promp… | High | — | 2026-09-26 |
| GHSA-v234-4jrq-mgg6 | Claude Desktop | Claude Desktop (macOS): opening a malicious file from a Cowork folder could run commands on the host | High | 1.15962.0 | 2026-09-25 |
| CVE-2026-49986 | Cortex MCP server | The Cortex MCP server (`neuro-cortex-memory`), a cross-platform persistent memory MCP, prior to version 3.17.1 treats the `CLAUDE_PROJECT_DI… | High | — | 2026-08-14 |
| CVE-2026-73614 | Network-AI ClaudeHookBridge | Network-AI ClaudeHookBridge before 5.15.1 truncates the target string to 500 characters before evaluating denyPatterns, while Claude Code ex… | High | — | 2026-08-13 |
| CVE-2026-73222 | Claude Code Templates | Claude Code Templates is a CLI tool for configuring and monitoring Claude Code. Prior to 1.29.4, the Claude Code Studio server launched by t… | High | — | 2026-08-11 |
| CVE-2026-57495 | AgenticMail | AgenticMail gives AI agents real email addresses and phone numbers. In @agenticmail/claudecode prior to version 0.2.39, @agenticmail/codex p… | High | — | 2026-07-20 |
| CVE-2026-47751 | Claude Code Action | Claude Code Action is a general-purpose GitHub action that runs Claude Code on GitHub pull requests and issues. Prior to 1.0.74, because the… | Medium | — | 2026-07-16 |
| CVE-2026-44467 | Claude Desktop | SSH Host Key Verification Bypass Allows Man-in-the-Middle Attack on Remote Sessions | High | 1.4304.0 | 2026-05-06 |
| CVE-2026-44470 | Claude Desktop | Local Privilege Escalation via Directory Junction in CoworkVMService | High | 1.3834.0 | 2026-05-06 |
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.
- Review
.claude/settings.jsonpermission rules in every cloned repo before trusting it.
Data exposure
What Claude Code 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 |
|---|---|
| Session transcripts in plaintext JSONL: messages, tool calls and tool results (file contents, command output, pasted text), plus the system prompt and loaded instruction files. | ~/.claude/projects/<project>/<session>.jsonl (subagent transcripts and large tool outputs in ~/.claude/projects/<project>/<session>/) |
| Prompt history you typed, with timestamp and project path. Not removed by the age-based cleanup. | ~/.claude/history.jsonl |
| Pre-edit snapshots of files Claude changed (checkpoint restore). | ~/.claude/file-history/<session>/ |
| Contents of large pastes. | ~/.claude/paste-cache/ |
| Per-session debug logs, only while debug logging is on. | ~/.claude/debug/ |
| Login credentials. | macOS Keychain (falls back to ~/.claude/.credentials.json, mode 0600, if the Keychain rejects the write); Linux ~/.claude/.credentials.json (0600); Windows %USERPROFILE%\.claude\.credentials.json |
Sent to the vendor
All prompts and model outputs go to the model provider over TLS 1.2+. Usage metrics exclude code, prompts and file paths; error reports are redacted. /feedback, /bug and /share send conversation history including code. WebFetch sends the requested hostname (not the URL) to api.anthropic.com for a safety check.
Retention and training
Consumer plans: 5 years if model-improvement is allowed, otherwise 30 days. Team, Enterprise and API: 30 days and no training on Claude Code data by default; zero data retention is available to qualified Enterprise accounts. /feedback transcripts are kept 5 years.
Settings that reduce exposure
cleanupPeriodDays- Sets how many days local transcripts and related session files are kept (default 30, minimum 1).
CLAUDE_CODE_SKIP_PROMPT_HISTORY=1- Skips writing transcripts and prompt history to disk; such sessions cannot be resumed.
--no-session-persistence(with-p)- Non-interactive runs do not persist the session transcript.
permissions.denyrules such asRead(./.env)- Blocks reads of credential files so their contents never enter a transcript.
claude purge <path>- Deletes one project's transcripts, memory, file history and matching
history.jsonllines after confirmation. DISABLE_TELEMETRY=1/DISABLE_ERROR_REPORTING=1- Turns off usage metrics / error reports.
DISABLE_FEEDBACK_COMMAND=1- Disables
/feedback, which uploads conversation history. CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC- Disables all non-essential traffic at once (metrics, error reports, surveys); does not affect the WebFetch check.
For administrators
- Managed settings file:
/Library/Application Support/ClaudeCode/managed-settings.json(macOS),/etc/claude-code/managed-settings.json(Linux/WSL),C:\Program Files\ClaudeCode\managed-settings.json(Windows), or MDM / server-managed settings. - A managed
cleanupPeriodDaysenforces the local transcript retention period on every machine; theretention_sweepOpenTelemetry event reports whether the sweep ran. forceLoginMethodandforceLoginOrgUUIDrestrict sign-in to the organization's accounts.allowManagedPermissionRulesOnly,allowManagedMcpServersOnlyandallowManagedHooksOnlymake managed policy the only source of permission rules, MCP servers and hooks.- OpenTelemetry export redacts prompt text and tool details by default;
OTEL_LOG_USER_PROMPTSandOTEL_LOG_TOOL_DETAILSmust be set to include them.
Sources: Claude Code docs: Data usage · Claude Code docs: The .claude directory (application data) · Claude Code docs: Identity and access management (credential storage) · Claude Code docs: Managed settings · Claude Code docs: Settings (permission rules) · Claude Code docs: Environment variables · Claude Code docs: Monitoring (OpenTelemetry) · 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/claude-code/check?version=<your-version>" Or ask your agent through the ThreatDeck MCP server (https://threatdeck.net/mcp):
check_ai_tool_version { "slug": "claude-code", "version": "<your-version>" } Updated 2026-11-22