AI POLICY · AI ACT ART. 50 · L. 132/2025

Cite or refuse. Signed register. Local at search time.

Lemnia refuses to answer when the source is missing. Every query leaves a trace in a BLAKE3-signed register, and every exported document carries a C2PA manifest and a European qualified timestamp (European Trust List, eIDAS), in compliance with AI Act art. 50 from 2 August 2026. Mandatory citation applies to all packs, both strict, such as reminder, quotation and expert report, and hedged, such as the customer dossier.

PRINCIPLE 01

Citation, or refusal.

Every sentence Lemnia produces is anchored to a source document. When the source is missing, Lemnia refuses to answer with the phrase "Non ho evidenza sufficiente per questa affermazione" and forms no hypothesis. The operator is informed that the absence of an answer is, in that case, the correct answer.

The cite-or-refuse pipeline runs in 5 steps: (1) decomposition of the answer into atomic claims; (2) substring match against the evidence set; (3) mDeBERTa-NLI entailment verification; (4) consistency check against the knowledge graph; (5) strip-and-replace for non-entailed claims.

PRINCIPLE 02

Every query and every answer leaves a signed trace.

Every interrogation enters an append-only processing register, sealed BLAKE3 per entry. Each field is captured and signed: timestamp, tenant id, operator id, query text, retrieval path, source citations, output text and model hash.

The register is exportable to PDF, JSON-LD and CSV, all signed. It satisfies the GDPR Art. 30 record-of-processing obligation and the evidentiary standard implied by Tribunale di Siracusa 338/2026 (Art. 96 c.p.c. gross negligence).

PRINCIPLE 03

Local at search time, with the cloud reached only on opt-in.

Query-time retrieval and generation run on the company's hardware. No fragment of customer, supplier or employee data leaves the LAN, and the local Italian-trained models (Qwen3.5-4B Q4_K_XL, mDeBERTa NLI, Qwen3-Embedding-0.6B) handle every interrogation.

Cloud-burst (Pro mode) is opt-in and runs only ingest and long-generation jobs. Each batch requires per-batch consent, captured in the signed register. The cloud worker uses EU-hosted infrastructure (RunPod EU, Hetzner SEV-SNP planned), with no US data transfer.

PRINCIPLE 04

Deterministic by default, with agentic loops never on local models.

The default execution mode is deterministic state machines with the LLM as a bounded tool: an approach that is predictable, fast, auditable and low in token cost. Voice walk-up, standard search, the curated reports, the deterministic drafts and the low-latency UX paths all stay strict deterministic.

Agentic LLM tool-use loops are permitted only when genuinely a better fit than a deterministic decomposition (multi-step ambiguous research, novel query shapes, no template match). The constraint is firm: agentic loops run only on cloud-side models (Qwen3.6-35B-A3B-FP8, Velvet-14B). Local Qwen3.5-4B agentic is banned, because local precision is insufficient for tool-use loops and produces unreliable, uninsurable behaviour.

PRINCIPLE 05

Company data does not train the Lemnia models.

Customer data is never used to train, fine-tune or evaluate the Lemnia models. The model pipeline is trained on public Italian corpora (Wikipedia IT, GovIT, ItaCorpora, Italian Constitutional Court rulings) and on synthetic data generated by Lemnia's own training pipeline.

Disambiguation decisions made inside a tenant's instance (HITL modal answers) feed continuous training data limited to that single instance. No cross-tenant data flow occurs, and no model improvement propagates across customers.

REGULATORY ANCHOR

Garante Provvedimento 474/2025 + Trib. Siracusa 338/2026.

Two Italian decisions define the defensible posture for generative AI in operational contexts: Garante Provv. 474/2025 (Careggi case, €80,000 fine for use of generative AI without verified sources on patient records), and Tribunale di Siracusa 338/2026 (Art. 96 c.p.c. gross negligence for citation of AI-generated case-law precedents).

Lemnia is built around the explicit interpretation that, in Italy, the only insurable posture is mandatory citation combined with query-time local execution and a signed register. This AI Policy makes that posture contractual.

FOUNDERS PROGRAMME · LIMITED PLACES

Lemnia running on the data of a real company.

A thirty-minute demonstration, calibrated to the company's sector. Lemnia composes the record of a real customer, cites the sources line by line and presents the signed register ready for the DPO.

Request a pilot→Download the technical dossier