Salvato in:
| Autore principale: | |
|---|---|
| Natura: | Recurso digital |
| Lingua: | inglese |
| Pubblicazione: |
Zenodo
2026
|
| Soggetti: | |
| Accesso online: | https://doi.org/10.5281/zenodo.19557371 |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
Sommario:
- <p>The Deterministic Autonomous Infrastructure Governance System (DAIGS) architecture was originally designed to govern massive, safety‑critical industrial environments—from nuclear fusion reactors to uncrewed aerospace platforms. However, the rapid proliferation of consumer‑facing Large Language Models (LLMs) and local agentic ecosystems introduces a structurally different class of adversarial threats: prompt injection, toxic content generation, unauthorized data exfiltration, and hallucination‑driven corruption of Personally Identifiable Information (PII). I present <strong>Lume‑Omni</strong>, a consumer and edge‑focused adaptation of the DAIGS architecture that proves deterministic governance invariants can be mathematically and efficiently enforced within browser sandboxes, WebAssembly (WASM) modules, and constrained mobile endpoints. By framing the consumer application state—including DOM boundaries, prompt envelopes, and data‑access layers—as a bounded control surface governed by Lume’s deterministic engine, I establish mathematically provable immunity against prompt‑driven exploitation. The architecture introduces six consumer‑specific state layers, eleven deterministic invariants, edge‑executed Ed25519 runtime certificates, PII exfiltration envelopes, and faceted personality synchronization that safely bridges micro‑consumer instances to the broader macroeconomic Trust Layer graph. Lume‑Omni bridges the mathematical rigor of industrial cyber‑physical safety with the high‑throughput, low‑latency requirements of everyday web interaction, proving that deterministic governance is not merely an industrial luxury but a mathematical necessity for the consumer internet.</p>