Security
A tool that reads your mail has to be careful with it. Here is how Antilure is built, and how to tell us when something is wrong.
Mail is processed on your computer only
The model runs inside the Antilure agent on your own machine. There is no cloud inference and no third-party AI service. The agent reads each message in memory and never writes it to disk.
The agent talks only to your own machine
The agent listens on your computer's loopback address and nowhere else. Every request needs a token. The Outlook add-in pairs with the agent using a six-digit code you read from the tray icon, with a lockout after repeated wrong codes. The Gmail extension reaches the agent through the browser's native messaging, so it never holds a token.
Signed model updates
Every model is signed. Before loading one, the agent checks the signature against keys built into it and the hash of every file. If any check fails, it keeps the model it already has.
What's stored is encrypted
Explanations and sender references in the agent's store are encrypted with a key held in your operating system's credential store. Message ids are kept as hashes.
The company console is yours
The console runs on your company's own server and receives metadata, not mail. It refuses records with any field beyond the agreed metadata. Passwords are stored as Argon2id hashes, repeated failed logins lock the account, and every admin change goes into an audit log. The console can't invent a model: agents accept only models signed with a key they already trust.
What Antilure doesn't protect against
Software already running as you on your computer can read or stop the agent. Antilure warns about messages; it doesn't replace keeping the computer itself clean, and it doesn't strip hidden instructions from what an AI assistant reads. The limits are listed on How it works.
Report a vulnerability
If you find a security problem in Antilure, the console or this website, please write to security contact address. Include what you found, how to reproduce it and the version you tested.
- We'll confirm we received your report and keep you updated while we fix it.
- Please give us a reasonable time to fix the problem before you publish it.
- Don't access other people's data, and don't test against systems you don't own.
to confirm: response times and an encryption key for reports