With all the talk about AI watermarks and new open models by Meta, it feels a bit futile to even try to announce the release of our very own first model series, but heck, let's shoot for it! ShinrAI v1.1 is online now – at mere 0.3B it provides highest-quality (f1>0.95) semantic encryption (not just masking) and PII protection based on the modernBERT architecture in real time. This is our internal model, which we built and used from 2024 to this date and which beats close to all existing PII-protecting models, even way above its size. huggingface.co/innovius/shin… This is just the start: two more models already finished training, and two more are currently running on the 4th largest Supercomputer, all to be released over the next days and weeks, each one much more capable and larger than the other. The goal is to take AI and privacy to a whole new level across all languages, all cultural contexts, and all use cases, including the very complex domains of finance, governance, medical, and… if it works (still testing hard) coding. We even built and released a multiplatform desktop App which allows you to just deploy the models locally and use them straight in Claude, Cursor, or ChatGPT. Right now, we did the most rigorous testing with Claude, so expect this to work best, but with time, all should work equally well. Get the App here at chtsafe.com/apps We will soon also share more of the training data, a paper with all details and of course the above mentioned models. This is also all part of our enterprise suite which runs this alongside our own harness as an on-premise service.

Tune in soon to start today live demos, presentations and moderations out of our Rostock Lunar Testlab 🚀 – at Berlin, Germany