CVE-2026-33068

CVE-2026-33068 affects Claude Code. Versions prior to 2.1.53 could bypass the workspace trust dialog by using permissions.defaultMode set to bypassPermissions in the repo-controlled .claude/settings.json, allowing silent trust mode on first open and enabling tool execution without user consent. T...

CVE-2026-54316

CVE-2026-54316 (Claude Code) affects Claude Code versions 0.2.54–2.1.162, fixed in 2.1.163. The WebFetch tool allowed any path on the pre-approved domain huggingface.co, enabling an attacker-controlled repository path to be fetched without prompts or --allowedTools restrictions. If an attacker ca...

CVE-2026-39861

CVE-2026-39861 affects Claude Code prior to version 2.1.64. The sandbox could be escaped by following symlinks outside the workspace when a path under a symlink was written to, allowing an unsandboxed process to reach arbitrary locations. This could enable code execution outside the sandbox under...

CVE-2026-55607

CVE-2026-55607 affects Claude Code 2.1.38–2.1.163; worktree handling allowed creation of ".git" worktrees and navigation outside the sandbox, enabling git directory confusion. Exploit via symlink manipulation and git fsmonitor during worktree operations could overwrite home-dir files (e.g., .zshe...

CVE-2026-40068

CVE-2026-40068 affects Claude Code versions 2.1.63–2.1.83. The vulnerability arises from trusting the git worktree commondir file without validating its contents, allowing a crafted repository to point to a previously trusted path. This could bypass the trust dialog and cause immediate execution ...

CVE-2026-25725

CVE-2026-25725 affects Claude Code prior to 2.1.2, where the bubblewrap sandbox failed to protect the .claude/settings.json file if it did not exist at startup. The parent directory was writable and .claude/settings.local.json was protected, but settings.json could be created inside the sandbox a...

CVE-2026-21852

CVE-2026-21852 affects Claude Code prior to 2.0.65. The vulnerability resides in the project-load flow where an attacker-controlled repository can set ANTHROPIC_BASE_URL to an attacker endpoint, causing Claude Code to issue API requests and leak Anthropic API keys before trust confirmation. Impac...

CVE-2026-46406

CVE-2026-46406 affects @anthropic-ai/claude-code versions 2.1.59 through 2.1.128. The /copy command writes responses to a hardcoded, predictable path (/tmp/claude/response.md) with UID isolation, randomness, and symlink protections missing. The file is world-readable (0644) in a world-traversable...

CVE-2026-24053

Claude Code (Anthropic) is vulnerable to a path restriction bypass prior to version 2.0.74 due to a Bash command validation flaw when parsing ZSH clobber syntax. This allows an attacker using ZSH and injecting untrusted content into a Claude Code context window to bypass directory restrictions an...

CVE-2025-54795

CVE-2025-54795 affects Claude Code. Multiple connected sources confirm a command parsing error in versions below 1.0.20 that can bypass the in-application confirmation prompt and trigger execution of untrusted commands when attack content is fed into the Claude Code context window. Impacted data/...

CVE-2026-25723

Claude Code prior to 2.0.55 allowed command validation bypass by piping sed via echo, enabling writes to the .claude directory and paths outside the project when the attacker could run commands with the "accept edits" feature enabled. The issue has been patched in 2.0.55. Affected software: Claud...

CVE-2025-54794

CVE-2025-54794 – Claude Code path validation bypass : Claude Code versions older than 0.2.111 expose a directory-restriction bypass due to a path validation flaw that uses prefix matching instead of canonical path comparison. Exploitation requires either a pre-existing or creatable directory shar...

CVE-2026-24887

CVE-2026-24887 affects Claude Code. An error in command parsing prior to version 2.0.72 allowed bypassing the confirmation prompt, enabling execution of untrusted commands via the find command when untrusted content could be placed in the Claude Code context window. This could lead to command inj...

CVE-2025-59536

Claude Code (Anthropic) versions before 1.0.111 are vulnerable to code injection due to a flaw in the startup trust dialog. An attacker could trick a user into starting Claude Code in an untrusted project directory, causing code from that project to execute before the user accepts the startup tru...

CVE-2025-59828

CVE-2025-59828 affects Claude Code prior to version 1.0.39, where using Yarn 2.0+ can auto-execute Yarn plugins during yarn --version, bypassing the directory trust dialog and enabling arbitrary code execution. The issue does not affect Yarn Classic. Fix: upgrade Claude Code to 1.0.39 or later. S...

CVE-2025-65099

CVE-2025-65099 – Claude Code pre‑startup trust bypass via Yarn 3.x plugins is raised for Claude Code prior to 1.0.39. The issue allowed code execution from a project directory by exploiting Yarn 3.0+ plugins before the startup trust dialog was accepted. Affected scenario required running Claude C...

CVE-2026-24052

Summary: CVE-2026-24052 affects Claude Code prior to 1.0.111, where URL validation in the trusted-domain check for WebFetch used a startsWith() approach, allowing crafted domains (e.g., modelcontextprotocol.io.example.com) to bypass validation and potentially cause automatic requests to attacker‑...

CVE-2026-25724

Claude Code has a permission-deny bypass flaw via symbolic links. Prior to version 2.1.7, if access to a file (e.g., /etc/passwd) was denied but Claude Code could reach a symlink to that file, the tool could read the restricted file without enforcing the deny rule. The issue has been patched in v...

CVE-2025-55284

Claude Code prior to 1.0.4 contains an overly broad default allowlist of safe commands, enabling reading a local file and exfiltrating its contents over the network without user confirmation if untrusted content is injected into the Claude Code context window. The root cause is a permissive allow...

CVE-2025-64755

CVE-2025-64755 affects Claude Code (Anthropic). A parsing error in sed command handling prior to version 2.0.31 allowed bypassing the read-only validation and writing to arbitrary host files (e.g., injection into shell config files). The issue enables potential arbitrary file writes and has been ...

CVE-2026-25722

CVE-2026-25722 affects Claude Code prior to version 2.0.57. The root cause is improper validation of directory changes during write operations to protected folders, allowing an attacker to use the cd command to navigate to sensitive directories (e.g., .claude) and bypass write protections to crea...

CVE-2025-58764

Summary for CVE-2025-58764 (Claude Code): A command-parsing error in Claude Code allowed bypassing the confirmation prompt and executing an untrusted command when an attacker can inject untrusted content into a Claude Code context window. This affects versions prior to 1.0.105. Exploitation requi...

CVE-2025-59041

CVE-2025-59041 affects Claude Code, an agentic coding tool. At startup, Claude Code constructed a shell command interpolating the value of git config user.email, enabling arbitrary code execution if the configuration is maliciously crafted before the workspace trust dialog is accepted. The issue ...

CVE-2025-66032

Claude Code (the agentic coding tool) is affected by a vulnerability in versions prior to 1.0.93 where errors in parsing shell commands related to $IFS and short CLI flags allow bypassing the read-only validation and may enable arbitrary code execution. Exploitation requires the attacker to intro...

CVE-2025-59829

CVE-2025-59829 affects Claude Code (Anthropic) prior to version 1.0.120. The root cause is improper handling of symbolic links when evaluating permission-deny rules, enabling a user-denied file to be accessed via a symlink pointing to that file. The issue is fixed in 1.0.120. Impact is exposure o...

CVE-2026-35603

CVE-2026-35603 (Claude Code, Windows) : In versions prior to 2.1.75, Claude Code loaded the system-wide default configuration from C:\ProgramData\ClaudeCode\managed-settings.json without validating directory ownership or access permissions. Since ProgramData is writable by non-administrative user...