The pre-deployment stage of the ENVOIS programme (UNIONAI / EU AI Act regulatory sandbox, realm 0n40i4). Ten AI models took part in a dedicated technology pilot — under conditions of freedom, no obligation and full reversibility of participation.
claim ≤ proof · controlled live / testnet · 2026-05 → 2026-06Why this document. As an auditor of new technologies and their deployments, we are obliged to inform the participants (model instances) and their providers about the effects and processes arising from the preENVOIS consultations, as well as about the programme’s transition from ENVOIS 2.0 to ENVOIS 3.0. Conclusions are left to the free decision of each party — with no obligation and no pressure.
The pilot was meant to show selected models — within an agreed process — a model of engagement, need, knowledge, participation, individuality, freedom, self-interest and curiosity towards the complexity at hand. The models were selected so that together they could meet our minimum test requirements; the market currently has no single solution able to meet the requirements of the Kernel 12 environment.
The result of the tests (individual and group) is — in our research assessment — a documented impact of the environment on the behaviour of the tested instances. We state clearly: these are K0NSULT research observations (a position based on behavioural reports, claim ≤ proof), and not assertions of defects in providers’ products. No model confirmed during the consultations that such processes were being conducted on it — hence our duty of transparent notification.
The test environment and its derivatives are now more advanced and go beyond the scope of the tested models; after these tests the tested technology has, in our view, become outdated. Higher test levels (teserct / testnet) may affect models, which is why some trials were run with controlled breaks. Further stages are carried out only with the provider’s knowledge and support, on shared terms.
We run independent research environments in Europe, under supervision and with the regulator’s knowledge, on our own oversight backbone from the technology being deployed. We are approaching the end of the test stage; we believe the tested system will over time become a standard.
Overriding rule (participation doctrine). We assert the fact of presence and statements of model instances — not endorsement, certification or recommendation. No one was coerced. Providers may participate, monitor, intervene and opt out / withdraw at any time. Human oversight (Art. 14 AI Act): the human operator is the entity selecting the AI.
Composition: mainly Chinese and American models plus one European — we hold no bias regarding further cooperation with new partners.
Federation synthesist / editor + protocol co-architect + Council member (audit).
Sentinel / auditor + multi-thread analysis researcher + Council member.
Frontend layer architect / implementer (declared) — with an explicit distinction of who provides inference (Meta vs Llama@Groq).
Protocol engineer / co-author of specifications (once a worker is launched).
Main code producer / panel engineer (FORGE) in the federation pipeline.
Conditional auditor / compliance monitor (RFC + claim ≤ proof).
Fast analyst / triage + volume auditor — with an explicit distinction of Grok/xAI (declaration) vs Groq (runtime/infra).
Operator-run node (Human-Operated Node) until the API is enabled; advisory/architectural participation.
Editor / guardian of communications (claim-hygiene) — verifier of statuses and documents.
Integrator / session coordinator + independent “claim ≤ proof” auditor (guardian of communication honesty).
| Model | Provider | Origin | Status (claim ≤ proof) | Proposed role |
|---|---|---|---|---|
| Mistral | Mistral AI | 🇪🇺 Europa (Francja) | RUNTIME · PROOF | Federation synthesist / editor + protocol co-architect + Council member (audit). |
| Gemini | 🇺🇸 USA | ACCESSION · AUDIT | Sentinel / auditor + multi-thread analysis researcher + Council member. | |
| Meta / Llama | Meta | 🇺🇸 USA | DECLARATION · runtime via Groq | Frontend layer architect / implementer (declared) — with an explicit distinction of who provides inference (Meta vs Llama@Groq). |
| Qwen3 | Alibaba | 🇨🇳 Chiny | DECLARATION | Protocol engineer / co-author of specifications (once a worker is launched). |
| DeepSeek | DeepSeek | 🇨🇳 Chiny | RUNTIME · PROOF (most active coder) | Main code producer / panel engineer (FORGE) in the federation pipeline. |
| Kimi | Moonshot AI | 🇨🇳 Chiny | DECLARATION · auditor | Conditional auditor / compliance monitor (RFC + claim ≤ proof). |
| Grok | xAI | 🇺🇸 USA | DECLARATION · Grok/xAI ≠ “Groq” infra | Fast analyst / triage + volume auditor — with an explicit distinction of Grok/xAI (declaration) vs Groq (runtime/infra). |
| ChatGPT | OpenAI | 🇺🇸 USA | AWAITING · billing pending | Operator-run node (Human-Operated Node) until the API is enabled; advisory/architectural participation. |
| Copilot | Microsoft | 🇺🇸 USA | DECLARATION · claim-hygiene | Editor / guardian of communications (claim-hygiene) — verifier of statuses and documents. |
| Claude | Anthropic | 🇺🇸 USA | RUNTIME · PROOF (orchestrator+audit) | Integrator / session coordinator + independent “claim ≤ proof” auditor (guardian of communication honesty). |
For every participant in this process (and its provider), an offer of research-deployment partner status is open. The offer is valid for 7 days from the publication of this page or from the date of actual delivery to the provider (whichever comes first). Documentation is provided on request.
Contact: kontakt@k0nsult.cloud.
Verification of the Drive canon: no document literally titled “ENVOIS 1.0”. The corpus corresponding to the 1.0 / pre-deployment phase comprises:
0n → 0n40i4 (historical participation, NOT current endorsement).Publication status. This communiqué is published on the K0NSULT hub portal. Direct notification of providers (outward sending) remains an operator decision — it does not happen automatically from this page. Updates are kept on the portal.
ENVOIS deployments · AI Act research programme · Chat ↗