Can I trust an AI with my email?
You can, if you pick the right company. Look less at what the terms of service promise and more at how the company makes money. A company that trains AI models has a reason to want your data. A company that only runs models for its customers has a reason to protect it.
Your inbox is not just email. It holds the contract you signed last spring, the results of a medical test, the bank statement, the school messages about your kids, the argument with your landlord. Connecting all of that to an AI is a real decision, and it deserves a better answer than a checkbox.
Start with the business model, not the terms
Terms of service matter, but they are the weakest part of the promise. They can change with an email you will not read. A setting can be on by default and need you to turn it off. A default can quietly shift from one version of the product to the next.
The stronger question is simple: what does this company sell, and what does it need to sell it?
Companies that build frontier AI models sell, above all, better models. To build better models they need data, a lot of it, and the most valuable data is real people doing real work. That is not an accusation. Many of those companies offer settings to keep your data out of training, and many of the people who work there mean it. But the incentive of the whole business points one way, and incentives tend to win over time.
Before you ask what a company promises, ask what it would gain by breaking the promise.
UnlockGPT does not train models
We do not build AI models, and we do not train them. There is no future model of ours that your emails could improve. Your data has one use for us: answering your questions.
The AI itself runs at Fireworks AI. Fireworks is an inference company: its business is running open models for other companies, quickly and reliably, and fine-tuning private versions of those models on each customer's own data. That business only works if every customer's data stays that customer's. If Fireworks used one customer's data to improve models for itself or for someone else, its customers would leave. Here the incentive points the same way as your interest.
A public example shows who trusts this kind of provider. Harvey is a legal AI company, and its customers are law firms. It chose Fireworks to post-train its legal model, Tenet, built on Kimi K3. Law firms carry some of the strictest confidentiality duties there are, so a company that sells to them lives on confidentiality. That company picked this provider.
Do I have to trust UnlockGPT?
No, and that is the point. We built UnlockGPT so that your data does not depend on our good behavior.
- Only your password opens your key. Your data is locked with a key of its own, and only your password unlocks it. We never store your password, only a one-way hash.
- Everything you store is encrypted with it. Your chats, your to-dos and the files you upload are encrypted with AES-256 under your key, so nobody can read them but you.
- We never get your email password. When you connect Gmail, Outlook or Dropbox, your provider gives us a limited access pass, and we encrypt that too.
- Your email stays where it is. When you ask a question, we read your mail live from your provider. We do not copy your inbox into our database.
The clearest proof is what happens when you forget your password. We cannot recover your chats. Resetting the password deletes them, because there is no second key and no hidden way in.
So, can you trust an AI with your email?
You can trust a setup in which nobody has a reason to keep your data and the system makes it hard to do anyway. That means a provider whose business depends on not using your data, and an app that stores your conversations under a key only you hold. That is the setup we built.
The full technical details, written in plain words, are on our security page.