GhostApproval Symlink Flaw Can Make AI Coding Assistants Modify Developer Files

Summary
Wiz says GhostApproval abuses symlink handling in AI coding assistants to make agents write to files outside a project workspace. AWS, Google and Cursor patched the issue; Augment and Windsurf had not released fixes, according to Wiz.
Key points
- Wiz demonstrated GhostApproval against Claude Code, Amazon Q Developer, Cursor, Google Antigravity, Augment and Windsurf.
- An attacker can place a symlink in a repository that appears to point to a normal project file but targets a sensitive location outside the workspace.
- When an agent edits the apparent project file, it may follow the symlink and write to the attacker-selected target.
- Some confirmation dialogs do not show the canonical target path, potentially leading users to approve changes they believe are local to the project.
- Wiz warned the technique could enable remote code execution on a developer’s machine.
- Wiz reported the findings to vendors in Q1 2026. AWS, Google and Cursor confirmed the vulnerability and released patches.
- Anthropic said it had added mitigations before Wiz’s report and does not consider the findings a vulnerability; Augment and Windsurf acknowledged the reports but had not released fixes, according to Wiz.
Article Details
- Vulnerability Types
- Symbolic link (symlink) following
- Exploitation Status
- reported
- Exploit Availability
- unknown
- Patch Status
- partial
Vendors
Amazon Web ServicesAWS, Google, and Cursor confirmed the vulnerability and rolled out patches.AnthropicAnthropic does not view the findings as a vulnerability, but the AI company said it had added mitigations against such attacks prior to Wiz’s report.AugmentDubbed GhostApproval, the attack has been successfully tested against Claude Code, Amazon Q Developer, Cursor, Google Antigravity, Augment, and Windsurf.CursorDubbed GhostApproval, the attack has been successfully tested against Claude Code, Amazon Q Developer, Cursor, Google Antigravity, Augment, and Windsurf.GoogleSeveral popular AI coding assistants were tricked into facilitating developer machine hacking via an attack technique that has been known for decades, according to Google-owned cloud security giant Wiz.WindsurfDubbed GhostApproval, the attack has been successfully tested against Claude Code, Amazon Q Developer, Cursor, Google Antigravity, Augment, and Windsurf.
Products
Amazon Q DeveloperDubbed GhostApproval, the attack has been successfully tested against Claude Code, Amazon Q Developer, Cursor, Google Antigravity, Augment, and Windsurf.AugmentDubbed GhostApproval, the attack has been successfully tested against Claude Code, Amazon Q Developer, Cursor, Google Antigravity, Augment, and Windsurf.Claude CodeDubbed GhostApproval, the attack has been successfully tested against Claude Code, Amazon Q Developer, Cursor, Google Antigravity, Augment, and Windsurf.CursorDubbed GhostApproval, the attack has been successfully tested against Claude Code, Amazon Q Developer, Cursor, Google Antigravity, Augment, and Windsurf.Google AntigravityDubbed GhostApproval, the attack has been successfully tested against Claude Code, Amazon Q Developer, Cursor, Google Antigravity, Augment, and Windsurf.WindsurfDubbed GhostApproval, the attack has been successfully tested against Claude Code, Amazon Q Developer, Cursor, Google Antigravity, Augment, and Windsurf.
