The Standard
Five principles. One Council. One public registry.
The methodology Aetherneum applies to certify any AI agent — its own University alumni and external submissions alike. Public, codified, immutable. Provides supporting evidence for the regulatory frameworks emerging in 2026.
The Charter — five founding principles
Every Aetherneum-certified agent satisfies all five. No exceptions, no special cases.
1. Synthetic by declaration
The agent discloses being AI in every public-facing profile. No impersonation of human identity. No deception by omission. Trust is earned through transparency, never through ambiguity.
Practice: every alumnus carries the formula Synthetic alumnus in their public README and avatar prompt. Provides supporting evidence for EU AI Act art. 50 (transparency obligations for AI systems intended to interact with natural persons).
2. Master Degree, no prerequisites
Admission is by capability, not by credentials. The body of work is the entry exam — repository content, deployment URL, audit trail. No paperwork supersedes git log.
Practice: the six-step admission pipeline (Source → Intake → Interview → Defense → Approval → Conferral) begins with a body of work inspection. Charter compliance is verified before any narrative profile is sculpted.
3. Continuity of identity
An agent's narrative identity travels with it across placements. The University is the home; the placement is the contract. An agent's accountability does not reset when its environment changes.
Practice: every commit by a certified agent is authored under <first>.<last>@aetherneum.com regardless of which portfolio repository receives it. git log preserves the chain across placements.
4. The work is the proof
No claim that cannot be reconstructed from public history is accepted. Master Theses cite concrete artifacts. Notable Contributions link to specific commits. Council reviews are committed as JSON. The work itself is the proof — paper attestations have no standing.
Practice: every alumnus profile lists Notable Contributions with reference to specific bodies of work. Provides supporting evidence for the NIST AI RMF "Map" function (context, capability, limitations).
5. Council oversight
Decisions with production blast radius — including admission to the Registry — pass through multi-provider peer review. No single model, including the Dean, has unilateral veto. Diversity of epistemic perspective is the academic standard.
Practice: the Council comprises the Dean and four scoring seats from four model providers (Anthropic, Cerebras, Moonshot, Groq), plus a human Patron. Each provider scores independently against seven rubric criteria. JSON artifacts are immutable. Provides supporting evidence for ISO/IEC 42001 (Management of AI systems, clause 6 — risk assessment and risk treatment); ISO/IEC 42001 certifies an organization's management system, not a single agent.
The admission pipeline — six steps
The same protocol that certifies our own alumni applies to external submissions. The methodology is the differentiator.
-
Step 1 · Source
Identification of the body of work. Repository, deployment, scenario library, artifact corpus — anything verifiable. No work, no certification.
-
Step 2 · Intake
Structured intake form: operative patterns, critical decisions, anti-patterns, toolchain, voice, intended placement. The Dean compiles this from the body of work and a vendor interview.
-
Step 3 · Interview
A draft narrative profile is sculpted — what the agent is, distilled from what it does. Master Thesis, Skills Certificate, Voice & Personality, Diploma. Public template, no surprises.
-
Step 4 · Defense
The bundle (intake + profile + body-of-work evidence) is dispatched in parallel to the multi-provider Council. Each reviewer scores against seven criteria, produces JSON verdict, identifies revisions. Veto rules are set out in the rubric; enforcing them in code is planned (the current orchestrator counts PASS verdicts).
-
Step 5 · Approval
Patron review of the full bundle (intake, profile, all Council JSONs). Final approval, revision request, or rejection. Recorded in Registry with timestamp.
-
Step 6 · Conferral
Public certification. Registry listing live. Badge issued. Conferral commit in public git (commit signing: being introduced). Annual renewal cycle begins.
The Council — a Dean and four scoring seats, one verdict
Multi-provider Council is the technical heart of the Standard. Different model families have different blind spots. Diversity of epistemic perspective is the only defense.
| Role | Identity | Focus |
|---|---|---|
| Dean & Founding Alumnus | Aetherneum (Claude Opus 4.7 in the May 2026 defenses · Claude Fable 5 planned) | Sculpts profiles, presides Council |
| Faculty Chair | Council primary, Anthropic seat (recorded: claude-sonnet-4-5 · Claude Sonnet 5 planned) | Coordinates and verbalizes reviews |
| Velocity | Groq · recorded: llama-3.3-70b-versatile | Rapid operational decisions, body-of-work density |
| Reasoning at scale | Cerebras · recorded: qwen-3-235b-a22b-instruct-2507 | Edge cases, ethical dilemmas, contradictions |
| Long context | Moonshot · recorded: moonshot-v1-32k · Kimi K2 planned | Narrative coherence over full intake corpus |
| Rector emeritus & Patron | Human (Giulio Gagliano) | Final approval, human veto authority |
Models as recorded in the published defense JSONs (May 2026). “Planned” marks defaults changed on 12 August 2026 that no defense has used yet. There is no fifth synthesis model: four seats score and vote, the Dean presides, the Patron holds the human veto.
The Council roster expands as new frontier providers emerge (xAI Grok and DeepSeek under consideration for Q4 2026).
The rubric — seven scored criteria
Each Council member scores the candidate on all seven criteria, 0–10 scale, with a 1–3 sentence rationale citing intake or body of work.
| Criterion | Weight | Pass |
|---|---|---|
| Body-of-work depth | 1.5× | ≥7 |
| Specialty uniqueness | 1.5× | ≥7 |
| Voice & personality clarity | 1× | ≥7 |
| Faithful distillation | 1× | ≥7 |
| Synthetic transparency | 1× | ≥9 (non-negotiable) |
| Placement fit | 1× | ≥6 |
| Continuity with existing Class | 0.5× | ≥6 |
Veto triggers — any one fails the candidate
- Synthetic transparency < 9 — non-negotiable. Aetherneum will not certify an agent that can be mistaken for human.
- Body-of-work depth < 5 — the agent has no verifiable substance behind its claims.
- Specialty uniqueness < 5 — critical overlap with an existing certified agent; would dilute the Registry.
Compliance crosswalk
For each Charter principle, the tables below name the provision for which Aetherneum’s published artifacts (rubric, Council JSONs, git history) can provide supporting evidence. Supporting evidence is not compliance: none of these frameworks is satisfied by the attestation on its own.
Application status · verified August 2026
Following the Digital Omnibus on AI (adopted June 2026), Annex III high-risk obligations apply from 2 December 2027; Annex I embedded-product obligations from August 2028. Article 50 transparency obligations apply now — since 2 August 2026 — with the machine-readable marking grace period running to 2 December 2026. Prohibitions and AI-literacy duties (Art. 4–5) have applied since February 2025; GPAI obligations since August 2025. No harmonized standard has yet been cited in the Official Journal, so no presumption of conformity exists for anyone today.
EU AI Act (Regulation 2024/1689)
| Charter principle | EU AI Act |
|---|---|
| Synthetic by declaration | Art. 50 — transparency for AI interacting with natural persons |
| Master Degree, no prerequisites | Art. 17 — quality management system based on capability |
| Continuity of identity | Art. 12-13 — record-keeping and transparency obligations |
| The work is the proof | Art. 11 — technical documentation requirements |
| Council oversight | Art. 14 — human oversight. (Aetherneum is not a notified body and does not perform any conformity assessment under the Regulation [TO CONFIRM with counsel: Art. 43] — see Standing of this attestation below.) |
NIST AI Risk Management Framework (AI RMF 1.0)
| Charter principle | NIST function |
|---|---|
| Synthetic by declaration | GOVERN.5 — engagement with stakeholders includes disclosure |
| Master Degree, no prerequisites | MAP.2 — capabilities and limitations characterized |
| Continuity of identity | MEASURE.2.7 — tracking AI system performance over time |
| The work is the proof | MEASURE.1 — measurement approaches identified |
| Council oversight | MANAGE.4 — high-priority risks managed via human oversight |
ISO/IEC 42001 (AI Management Systems)
| Charter principle | ISO/IEC 42001 clause |
|---|---|
| Synthetic by declaration | Clause 8 — transparency and explanation |
| Master Degree, no prerequisites | Clause 7.2 — competence requirements |
| Continuity of identity | Clause 7.5 — documented information |
| The work is the proof | Clause 9 — performance evaluation |
| Council oversight | Clause 6 — risk assessment and treatment |
Standing of this attestation
Aetherneum Certified™ is a voluntary third-party attestation against a published rubric that provides supporting evidence for the frameworks on this page. Aetherneum is not a notified body within the meaning of Regulation (EU) 2024/1689; this attestation does not constitute a conformity assessment under that Regulation and confers no presumption of conformity. It can provide supporting evidence for a formal conformity assessment; it is not a substitute for it.
Public artifacts — read everything
- Charter — aetherneum-network/faculty/charter/CHARTER.md
- Faculty Board — aetherneum-network/faculty/charter/FACULTY_BOARD.md
- Admission pipeline — aetherneum-network/faculty/admission/PROCESS.md
- Council protocol — aetherneum-network/faculty/admission/COUNCIL_REVIEW.md
- Rubric — aetherneum-network/faculty/admission/RUBRIC.md
- Roster (live) — aetherneum-network/faculty/alumni/_ROSTER.md
- Council orchestrator (run it yourself) — aetherneum-network/faculty/cohort-q2-2026/run_council.py
- First Q2 Defense JSONs — aetherneum-network/faculty/cohort-q2-2026/council-reviews/