Intellectual Property
Application No. P202631047 · OEPM · Filed 19 July 2026
Application number
P202631047
Filing office
OEPM — Oficina Española de Patentes y Marcas
Priority date
19 July 2026
Titleholder
Alan Daniel Lapid
PCT / EPO deadline
19 July 2027 — Global filing window open
Category
Privacy-native memory encoding for LLM inference pipelines
What was invented
The patent protects two independent inventions that together make privacy in AI memory a mathematical property rather than a policy choice. Neither invention requires model fine-tuning. Neither transmits sensitive data. Both operate before inference — not after.
IP · 01
Intra-session coherence with inter-session opacity. Each session is cryptographically isolated. The re-identification probability across sessions is bounded by 1/k — provably, not by policy. No published equivalent exists in the academic or commercial literature.
Technical implementation details available under confirmed NDA
IP · 02
Personally identifiable information is transformed before it reaches any language model — without retraining the model, without transmitting an equivalence table, without any persistent mapping that could be reversed. Automatic rotation. Works on any LLM, local or remote.
Technical implementation details available under confirmed NDA
The distinction matters. When privacy is a policy, it can be changed, overridden, or breached. When privacy is a mathematical property of the data itself, it cannot be undone — only the mathematics can.
Both inventions in P202631047 treat privacy as architecture, not governance. The system is private because of how it is built, not because of what it promises.
This is the Geometrical commitment: local-first by design, private by mathematics, transparent in its constraints.
IP timeline
Supporting research
Personal Episodic Identity Substrate (PEIS): A 9-Dimensional Architecture for Human Memory in the Agent Economy
cs.AI · Geometrical Research · 2026 zenodo.org/records/21923601 Zenodo · PublishedThe AIX Protocol: A Machine-Native Format for Contextual Identity in Multi-Agent Systems
cs.AI · Geometrical Research · 2026 zenodo.org/records/21923809IP Technical Detail Paper — Full disclosure of the two inventions following PCT filing in 2027.
cs.CR + cs.AI · Geometrical Research · 2027 Available after global patent filing