The IP-Safe AI glossary
The words behind using AI without leaking your work, in plain English. If you have run into a term on this site or elsewhere and were not sure what it meant, start here.
Public disclosure
Making an invention available to the public before you file for a patent. It can cost you the right to patent the invention, because a patent protects something new. Sending an unfiled invention to an outside party, including an AI tool, can risk counting as a disclosure. Europe has no grace period, so a disclosure before filing can permanently bar a European patent.
Training data
The text and other content an AI model learns from. If a tool trains on your inputs, what you paste can become part of a future model and help produce answers for other people. Many consumer tools train on inputs by default unless you opt out.
Zero-knowledge
A property where a system can be useful to you without any single part of it holding your complete secret. IPVault applies the idea to AI: your document is split so no single AI model ever sees the whole, and the code that reverses the masking never leaves your device.
Masking
Replacing names, secrets, and identifying details with codes before anything is sent, so the AI sees PERSON_A instead of a real name. The code book that maps codes back to real values stays on your own device.
Opt-out
A setting that tells an AI tool not to use your inputs for training. It helps, but it is forward looking only, it relies on the company honoring it, and defaults can change. Opt-out is not the same as privacy.
On-device
Work that happens on your own computer and never leaves it. In IPVault, reading and understanding your document, masking it, and reassembling the result all happen on-device. Passages you mark as extra sensitive stay on-device and are never sent to any AI.
Attorney-client privilege
A protection that keeps confidential communications between a lawyer and client from being disclosed. It depends on keeping the communication confidential, so sharing privileged material with an outside party that does not need it, such as a consumer AI tool, can put the privilege at risk.
IP-Safe AI
Using AI on your work in a way that protects your intellectual property: no single AI sees the whole, nothing is used to train a model, and you can keep the most sensitive parts on your own device. It is the category IPVault is building.
Questions, answered
What does zero-knowledge AI mean?
It means using AI where no single model ever holds your complete work, and the information needed to reverse the masking never leaves your device. You get the help without any one system knowing the whole.
What is IP-Safe AI?
Using AI in a way that protects your intellectual property: no single model sees your whole document, nothing trains on your input, and you can keep your most sensitive passages on your own device.