Nakato is a private layer between your sensitive data and public AI models. Your real data never leaves your environment. The model works on a semantic twin, and you still get the accuracy you needed.
Synthetic example. No real customer data appears on this page.
See the swap in four sectorsHow it works
01 Semantic
Nakato reads the whole document and finds every sensitive detail — direct or indirect, including the indirectly identifying information that pattern matching misses.
02 Pseudonymisation
Every sensitive value is swapped for a semantic twin: fictional, but statistically faithful — same shape, same context, different person. The model only ever sees the twin.
03 Reversible
When the answer comes back, the swap is exactly reversed and every substitution is auditable. Your real data never leaves your environment.
| Type | Original | State |
|---|---|---|
| person | Margaret Ellis | found |
| date of birth | 1961-04-12 | found |
| account number | 20-45-11 88451203 | found |
| address | 14 Rowan Grove, Sheffield | found |
4 values found, read for meaning rather than matched.
It happens live, at the moment of each query — not as an up-front dataset transformation. The model gets the context it needs to reason, and it has no reasonable means of identifying your customers.
The demo
One layer, wherever regulated data is the blocker. Pick a sector: the record on the left stays in your environment, and the model reasons on the column beside it.
Credit decisions on the full picture
Credit decisions on the full picture. Strip demographics and you strip the vulnerability signals credit analysis depends on most. The twin keeps every signal and removes every identity, so the model reasons on complete records.
The model only ever saw the right-hand column; the answer you read had the originals restored. Where the two columns match, that is the point — the twin changes what identifies and keeps what the reasoning needs.
Evidence
The field is the benchmark. Each point stands for the same number of test cases. Each mark reaches as far as that tool's figure.
Summarisation accuracy on the same cases. A longer bar is better.
Shown alone. A comparison is drawn only where the same benchmark produced one.
Shown alone. A comparison is drawn only where the same benchmark produced one.
Leak rate under the same attack. A shorter bar is better.
Checked across 500,000 financial-services test cases; every benchmark is published for review.
Sign-off waits on a familiar set of questions. The short answers are here; the particulars — and the frameworks the layer is built against — are on their own page.
Where does our data go?
The model never receives a real value, and the transformation never leaves your infrastructure — so nothing about your existing data classification has to change.
What happens when detection is unsure?
An entity the detector is not certain about is held back rather than passed through. The layer is designed so the model has no reasonable means of identifying your customers.
Can we evidence what was substituted?
Every swap is written down — what was replaced, with what, and when — so a reviewer can retrace a decision without ever seeing the underlying data.
Built against the EU AI Act, DORA, FCA and PRA expectations, GDPR and SOC 2 from the start.
Join us
Generating a twin that keeps its meaning is not a solved problem. It takes an unusual mix of semantic ontology and computer science, and no two domains ask the same thing of it.
Semantic ontology meets computer science. Two disciplines, one twin that keeps its meaning.
The problem changes shape with every domain. What preservation means for medical data looks nothing like insurance.
The work does not repeat. Small team, hard problems, and no template to follow.
We intend to be the ones who define how this is done.
Your real data never leaves your environment. The model only ever sees the twin, and you still get the accuracy you needed.