WIPO Publishes Foundational AI System Architecture by an Independent Italian Researcher

Bielefeld, 22 August 2026. While the public AI debate is increasingly focusing on reflection, self-correction, drift, stable context management, and governance, an international patent application has been publicly accessible since 20 August 2026 that already addressed key questions of this development architecturally in February 2025.

The independent Italian researcher Salvatore Orto from Bielefeld developed the underlying system architecture largely outside the public AI debate. The development established priority on 14 February 2025 with U.S. Provisional Application 63/758,361 and was filed internationally on 21 February 2025 as PCT/IB2025/051893. The World Intellectual Property Organization (WIPO) has now published the application under WO 2026/172133. This was followed on 11 February 2026 by U.S. Non-Provisional Application 19/xxxxxx.

The architecture describes neither a new language model nor a single AI feature. It addresses a more fundamental question: What must happen within a digital system before an impulse becomes a response or an action at all?

Instead of inevitably translating an input into output, the architecture routes impulses through interconnected reflection and evaluation spaces. Different response paths can remain open simultaneously, while possible reactions are fed back and evaluated again. A higher-level control structure coordinates these processes. The system can carry an impulse forward, feed it back, or hold it, delay a reaction, or refrain from reacting altogether.

The technical capability to react therefore does not automatically constitute authorization to react.

Between impulse and action, a structural space emerges in which uncertainty, contradiction, and alternative development paths can remain open. The architecture combines this with mechanisms for feedback, coherence preservation, self-correction, controlled non-response, and the handling of drift. Its components are not designed as functions that can be used in isolation, but as an interacting, non-fragmentable architecture.

Orto describes the approach as a Foundational and Enabling Architecture: not as an additional capability of an existing AI system, but as an architecture that defines the structural conditions for when and under what prerequisites downstream capabilities can become effective at all.

This development context gave rise, as early as spring 2025, to ORCA – Orto Resonance Carrier Architecture, in which carrier, resonance, and tension mechanisms in particular were further differentiated. The subsequent U.S. Non-Provisional formalizes key elements of the architecture, including through Carrier, Holding, Delay, Audit, and Non-Response structures.

The timeline of the development makes the publication particularly relevant. Numerous questions that are currently being addressed individually in the AI industry under different terms can now be compared with an architecture whose development was already documented through patent filings in early 2025.

Whether and to what extent later developments exhibit structural correspondences with this previously documented architecture is the subject of an audit and research documentation process initiated in April 2025 and continuously maintained since then. This provides not merely a retrospective assessment, but a documentation and comparison basis developed contemporaneously over time. With WO 2026/172133, the underlying chronological and technical starting point is now also independently and publicly verifiable for the first time.

Patent documentation: WIPO PATENTSCOPE – WO 2026/172133

International Application: PCT/IB2025/051893

Priority: US 63/758,361 – 14 February 2025

U.S. Non-Provisional: 19/xxxxxx – 11 February 2026

Scientific documentation:

Orto, S. (2026). Responsibility as Architecture: Foresight, Architecture, Holding and Silence – An Auditable System for the Control of Generative Processes under Drift Conditions (Version 1.1 Platform-Compatible Distribution English Version). Zenodo. https://doi.org/10.5281/zenodo.22159206