Security
Last updated October 8, 2026
Connecting an AI client to a terminal is powerful and carries real risk. Here is what the UnliShell service does and what you should do to stay safer.
Current safeguards
- Device pairing requires approval from a signed-in account, and paired devices can be revoked.
- AI clients connect through OAuth authorization. The server checks the account, authorized device and the requested tool permission before dispatch.
- Published HTTP endpoints use HTTPS. The service stores password verifiers and most device/session credentials as hashes instead of plain-text values.
- The activity database records tool names and operational metadata rather than intentionally retaining remote command and file output.
- Account owners can terminate device sessions and disconnect AI clients from the dashboard.
These controls reduce risk but do not make remote execution safe in every situation, and they do not mean the service has undergone an independent security audit.
Protect your computer
- Run the agent as a non-administrator, least-privilege user and restrict which folders that user can access.
- Keep independent backups before authorizing file writes, deletes, installations or system changes.
- Do not expose secret files, environment variables, browser profiles, payment data, medical records or identity documents to AI tools.
- Approve only AI clients you trust, examine requested read/write scopes, and revoke permissions immediately if anything looks wrong.
- Stop the client when you are not using it; secure your device and operating system with normal updates and endpoint protection.
- Review commands before permitting destructive, irreversible or externally transmitted actions. Prompts and untrusted file contents can mislead an assistant.
Report a vulnerability
Send a minimal, non-sensitive description to [email protected]. Include the affected component, steps to reproduce on your own account/device and potential impact. Do not access another person's account or device, exfiltrate data, disrupt service, or publicly share exploit details before coordinated disclosure. No bug bounty or reward is promised.
Emergency response
If you suspect compromise, disconnect your computer agent, revoke devices and AI connections, change your password, and secure the affected operating-system accounts. Contact support with a brief report. If any legal reporting deadline applies to an incident, we will follow applicable law.