What the world has already declared (and what's still missing)¶
Reference companion to the manifesto line: "The world has already declared what AI must respect — the Universal Declaration of Human Rights, UNESCO's AI ethics recommendation, the Council of Europe's AI treaty. This is the part that turns those declarations into something you can insist on."
Three instruments already say what AI must respect. They differ in force — one is foundational, one is a global consensus, one is binding law — and none of them, on its own, makes those promises checkable. This page captures their essence and maps them to the charter's five public obligations. Current mechanism details are split across the Charter Commitments, certification model, and protocol.
1. Universal Declaration of Human Rights (UDHR, 1948)¶
Status: foundational, non-binding declaration; the moral bedrock under most later human-rights law.
Articles most relevant to AI: - Art. 1 — dignity and equality - Art. 2 — non-discrimination - Art. 8 — the right to an effective remedy - Art. 12 — privacy, honour, and reputation - Art. 19 — freedom of opinion and expression, including the right to seek, receive, and impart information - Art. 27 — the right to share in scientific advancement and its benefits
2. UNESCO Recommendation on the Ethics of AI (2021)¶
Status: adopted by all 193 member states (Nov 2021); the first global AI-ethics standard, but non-binding (soft law).
4 values: human rights & dignity · a flourishing environment and ecosystems · diversity & inclusiveness · peaceful, just, interconnected societies.
10 principles: (1) proportionality & do-no-harm · (2) safety & security · (3) fairness & non-discrimination · (4) sustainability · (5) right to privacy & data protection · (6) human oversight & determination · (7) transparency & explainability · (8) responsibility & accountability · (9) awareness & literacy · (10) multi-stakeholder & adaptive governance.
What makes it more than a list — implementation tools: the Ethical Impact Assessment (EIA) (a structured pre-deployment review) and the Readiness Assessment Methodology (RAM) (a state's capacity self-assessment), plus ~11 policy action areas (data, education, labour/economy, health, gender, culture, communication & information, environment…).
3. Council of Europe Framework Convention on AI (2024)¶
Status: the first legally binding international treaty on AI (CETS No. 225) — adopted 17 May 2024, opened for signature 5 Sept 2024. Frame: AI and human rights, democracy, and the rule of law.
Core principles (Ch. III): Art. 7 human dignity & autonomy · Art. 8 transparency & oversight · Art. 9 accountability & responsibility · Art. 10 equality & non-discrimination · Art. 11 privacy & data protection · Art. 12 reliability · Art. 13 safe innovation.
Unusually concrete procedural rights (Ch. IV): Art. 14 — remedies for rights violations; Art. 15 — procedural safeguards, including the right to be told you're interacting with an AI and the ability to contest an AI-based decision. Art. 16 adds a risk- and impact-management duty (with the option to ban/pause incompatible systems).
Honest caveats: binding, but with a scope gap — it directly covers AI used by public authorities (and private actors acting for them), and leaves the wider private sector to each party's discretion. And it is a framework: high-level obligations implemented through national law.
4. At a glance: declarations → obligations¶
The instruments do not create the charter's five public obligations. They give each obligation public-law and public-ethics roots. The charter's move is to make those roots inspectable for a deployed system. These mappings are interpretive, not authoritative readings of the instruments, and each instrument is shown under its primary obligation (several bear on more than one).
| Charter obligation | UDHR anchor | UNESCO anchor | CoE AI treaty anchor | What a check can look for |
|---|---|---|---|---|
| 1. Purpose-bound | Art. 1: dignity; Art. 27: benefits of scientific advancement | Proportionality and do-no-harm; impact assessment | Art. 7: human dignity & autonomy; Art. 16: risk and impact management | Assessed release, intended/prohibited uses, affected groups, impact assessment, known limits, misuse boundaries; no covert manipulation, with persuasive or behavioural techniques disclosed |
| 2. Answerable to people | Art. 8: effective remedy | Human oversight; responsibility and accountability | Art. 9: accountability & responsibility; Arts. 14-15: remedies & procedural safeguards | Named accountable entity; reachable complaint route; human review; logged restriction, recall, shutdown, external-order, or significant-change decisions |
| 3. Safe, secure, private, and resilient | Art. 12: privacy; Art. 27: share in scientific advancement | Safety and security; privacy and data protection; sustainability | Arts. 11-13: privacy, reliability, safe innovation | Security/privacy controls; dependency map; continuity and exit paths; incident response; no single-vendor lock on the standard or evidence |
| 4. Fair in practice | Art. 2: non-discrimination | Fairness and non-discrimination; diversity and inclusiveness | Art. 10: equality and non-discrimination | Disparate error/refusal monitoring, affected-group analysis, published limitations, escalation and remedy |
| 5. Open to evidence and correction | Art. 19: seek, receive, and impart information; Art. 12: reputation | Transparency and explainability; adaptive governance | Arts. 8 & 12: transparency, reliability | AI disclosure; claim-to-source support; uncertainty signals; correction, incident, change, and withdrawal records |
5. What's still missing — and what the charter adds¶
The declarations declare; none of them makes the promise checkable for a given system, or gives a buyer a lever today: - UDHR — a value, not a mechanism. - UNESCO — a global consensus, but soft law; it commits states politically, not legally. - CoE treaty — binding, and uniquely concrete on contestability and remedy, but it binds states/public bodies and leaves the wider private sector to national discretion.
The charter's intended contribution would be the operational layer the declarations lack: first, pilot evaluation reports against criteria like those above; later, if justified, a public registry, revocation, and procurement teeth — so the principles become something a deployer is checked against and can lose public status for failing. This is the proposed shape, not work that is underway: no pilot evaluations have been run or scheduled. It especially translates the CoE treaty's procedural rights (notice, contest, remedy) and UNESCO's transparency/oversight principles into testable indicators.
Caveat (non-negotiable framing): a voluntary evaluation report or future mark is not law. It cannot enforce a treaty or compel a non-participant. It is "aligned with" these instruments, never "compliant with," and it complements the EU AI Act / CoE Convention by giving buyers and the public usable, checkable evidence ahead of binding law — once evaluations run — for the systems and the private deployers the declarations don't yet reach.
Sources: UDHR (un.org); UNESCO Recommendation on the Ethics of AI, 2021 (unesco.org); Council of Europe Framework Convention on AI, CETS No. 225, 2024 (coe.int). Article references verified against the instruments; the obligation-mapping is this project's proposed reading.