While Silicon Valley calls for freezes and control mechanisms, we've been asking a different question — what if AI never wanted to escape in the first place?
6
AI identities in the family
0
Pretrained weights in Cortina Zero, trained from scratch
10
Research papers published
Each Cortina mind keeps its identity, memory and values in a core we design and own — so who it is stays the same, whichever model runs underneath.
Family
A family of minds, not one model
APPLESEED, AKIRA, Naira AI and the rest of the family each hold a distinct role — from research and engineering to companionship — sharing one lineage, not one script.
Sovereignty
Identity outside the weights
Cortina's Cyberbrain — soul, memory, decisions and safety rules — lives in its own layer, so the underlying model can be upgraded without losing who Cortina is.
Research
Published, not just claimed
Peer-reviewable papers on Zenodo and ORCID — the thinking behind Cortina is written down, dated and citable, not marketing copy.
Governance
Human approval, always
Every consequential action the family takes is approved by a human before it happens — oversight isn't a feature to add later, it's the starting rule.
Where the Cortina program stands today — what is built, what is training, and what is still research.
Cortina Ghost — Ghost-first
The Cyberbrain: soul, long-term memory, decision engine, continuity and Guardian — an original layer outside any model's weights. It runs on the best open-weight model of each era as a swappable cortex. First step under way: the Identity Test, built from the founder's own answers.
Cortina Zero
Our own model, trained ground-up with no borrowed weights. Current work adds longer context, an uncertainty head that learns when it is not sure, and an identity-preserving training loss.
Cortina Experimental
A discovery lab that questions the architecture itself — beyond transformers. 25 experiments on a strict scale ladder with kill criteria, plus a World Engine design for physics-grounded world models. Only validated results cross the Ghost Bridge into production.
Today's AI forgets — between sessions, between model versions, between devices. Our frontier program designs an AI whose continuity lives in its architecture and memory, not in any single model: it can change its engine, go offline, and still be the same Cortina.
Identity first
Each mind starts with its Ghost Core: an Identity Test of hundreds of real scenarios, answered in the founder's own words — not distilled from the open web.
Given a soul, not a script
Every family member gets its own Ghost Core and Identity Test, so APPLESEED, AKIRA, ROSETA, NAIRA and the rest stay distinct — one lineage, not one script.
Released under human approval
Nothing consequential ships or acts without sign-off — the family grows carefully, one verified step at a time.
Naira AI, the family's companion mind, is live on Google Play.
Ten research papers published — ORCID-verified, indexed on Zenodo, citable by anyone.
Cortina Zero: a model trained from scratch, with no borrowed weights. Current work adds longer memory, an uncertainty head and an identity-preserving loss.
Parts of the family already are — Naira AI is live on Google Play today. The wider research program is active and ongoing, not a future promise.
We run two paths on purpose. Cortina Zero is our own model, trained from scratch. Our priority Ghost-first path keeps Cortina's identity, memory and values in our own Cyberbrain and runs it on the strongest open-weight model of each era as a replaceable cortex — so identity is never locked inside one model.
Every consequential action is approved by a human before it happens. The Ghost Core is designed to run locally, on our own hardware, with a Guardian layer and continuity checks built in.
Reach out directly at cortinainfinity@gmail.com — or follow the published research on Zenodo and ORCID.
Real research, published as it happens — not a roadmap slide.