Zero-Click 'SneakPeek' Flaw Silently Exfiltrates Developer Credentials from Claude Code
The 'SneakPeek' vulnerability in Anthropic's Claude Code Desktop exploits the application's preview pane to silently steal highly sensitive credentials — including SSH private keys, AWS tokens, and OAuth secrets — without requiring any user interaction. This zero-click attack vector is particularly dangerous because traditional security awareness training offers no defense; developers cannot 'avoid' a threat they never see or trigger. The exposure of SSH and cloud credentials can cascade into full infrastructure compromise, lateral movement, and persistent unauthorized access. This incident highlights the risk of storing long-lived, high-privilege credentials in environments adjacent to third-party tooling. Developers and security teams must treat AI-assisted developer tools with the same scrutiny applied to any privileged software in the software supply chain.
Tactical Insight
Immediate actions
- Apply the latest Anthropic Claude Code Desktop patch immediately and verify the updated version is deployed across all developer workstations.
- Rotate all potentially exposed credentials (SSH private keys, AWS access keys, Claude OAuth tokens) as an emergency precaution, even before confirming exploitation.
- Disable or sandbox the preview pane functionality in Claude Code until a verified fix is confirmed.
Long-term improvements
- Enforce short-lived, scoped credentials (e.g., AWS IAM roles with temporary STS tokens) rather than long-lived static keys stored on developer machines.
- Implement a secrets management solution (e.g., HashiCorp Vault, AWS Secrets Manager) to prevent credentials from residing in plaintext on endpoints.
- Establish a formal vendor security assessment process for all AI and developer productivity tools before organizational adoption.
Detection measures
- Deploy endpoint detection and response (EDR) tooling to alert on anomalous file access patterns targeting credential stores (e.g., `~/.ssh/`, `~/.aws/`).
- Enable cloud provider audit logging (AWS CloudTrail, etc.) and alert on credential use from unfamiliar IPs or geolocations.
- Integrate developer workstations into a SIEM to correlate suspicious outbound connections originating from IDE or AI assistant processes.