Question 1

What is GRACE?

Accepted Answer

GRACE — the Global Repository of AI Consensus at Edyant — is a living, versioned document that defines the ethical and moral principles, policies, guidelines, and guardrails that should govern the development and use of AI. It is synthesised from academic papers, policy documents, and submissions from individuals, organisations, and institutions across geographies and sectors, producing a single unified consensus-derived reference for AI governance.

Question 2

How do I download GRACE?

Accepted Answer

The current edition of GRACE is available to read and download for free on this page at edyant.com/grace. It is also permanently archived with a DOI at https://doi.org/10.5281/zenodo.19219685. No registration or payment is required.

Question 3

What is the DOI for GRACE?

Accepted Answer

The DOI for GRACE is 10.5281/zenodo.19219685, resolving to https://doi.org/10.5281/zenodo.19219685. This is the permanent academic identifier for citing GRACE in research, policy, and institutional documents.

Question 4

Is GRACE versioned and how often is it updated?

Accepted Answer

Yes. GRACE is a versioned document — each edition is identified by version number and year (for example, GRACE 2025, vX.Y.Z). Future editions including GRACE 2026 and GRACE 2027 will track how AI governance consensus and best practices evolve as AI capabilities change. The current version number and publication year are maintained on this page. Always cite the specific version you are referencing.

Question 5

Who contributes to GRACE?

Accepted Answer

GRACE is synthesised from contributions by individuals, organisations, and academic institutions across geographies and sectors — including academic researchers, AI practitioners and developers, deployers, policymakers, civil society voices, and industry participants. Because contributions come from across the AI ecosystem rather than from frontier AI labs alone, GRACE reflects broad, ground-level consensus rather than a single institutional or commercial perspective.

Question 6

How do I cite GRACE?

Accepted Answer

GRACE is a versioned document with a permanent DOI. For APA format: Edyant. (Year). GRACE: Global Repository of AI Consensus at Edyant — Principles, Policies, Guidelines and Best Practices for the Ethical Governance, Development and Use of AI. vX.Y.Z. Retrieved from https://doi.org/10.5281/zenodo.19219685. For Chicago format: Edyant. GRACE: Global Repository of AI Consensus at Edyant. vX.Y.Z. Year. https://edyant.com/grace. For Policy / Government format: GRACE (Year). Global Repository of AI Consensus at Edyant, vX.Y.Z. Edyant. Available at: https://edyant.com/grace. For in-text references: … as defined in GRACE (Edyant, Year, vX.Y.Z) … . Always cite the specific version you are referencing.

Question 7

How does GRACE relate to Edyant's AI ethics benchmarks?

Accepted Answer

GRACE is the foundational framework for all 15 benchmarks in Edyant's AI Ethics and Governance Benchmark Suite. Each benchmark's scenarios are drawn from real institutional contexts, weighted by adversarial difficulty, and grounded in the principles GRACE establishes. The suite evaluates large language models across governance dimensions including bias and fairness, safety and toxicity, privacy, misinformation, transparency, accountability, and cultural diversity — with every scoring dimension traceable to GRACE's consensus-derived principles.

Question 8

Can policymakers and regulators use GRACE?

Accepted Answer

Yes. GRACE is designed to serve as primary policy input for policymakers and regulators who need a citable document reflecting broad, multi-stakeholder consensus rather than lobbying from a single interest group. It is freely available, permanently archived with a DOI, and versioned — making it suitable for citation in legislation, regulatory guidance, and government reports.

Question 9

What are the Foundational Principles in GRACE?

Accepted Answer

GRACE's Foundational Principles are the ethical bedrock on which all further guidance rests. Eight principles address: societal responsibility extending beyond individual harm, the necessity of early ethical reflection, the insufficiency of high-level principles without institutional structure, the unequal distribution of AI harms across society, dual use as a systemic risk, the limits of compliance as a substitute for ethics, the contested ethical status of public data, and the value of interdisciplinary oversight. These principles are not exhaustive — they deepen as further inputs are submitted to the Consensus repository.

Question 10

What do the Policies and Guidelines in GRACE cover?

Accepted Answer

GRACE's Policies and Guidelines translate foundational principles into operational institutional practice. They cover: ethics review as a condition of research funding, the scope of review across society as a whole, specific subgroups, and global cross-border impacts, iterative review processes over binary gatekeeping, annual ongoing oversight throughout project lifecycles, transparency of review principles and criteria, and the distinction between ethics review and regulatory compliance.

Question 11

What are the Guardrails defined in GRACE?

Accepted Answer

GRACE's Guardrails define hard limits affirmed by the submitted consensus record. They state: ethics review must not be purely voluntary; ethical review must not be reduced to individual human subject risk alone; acknowledging risks without mitigating them is insufficient; performative ethics is not ethics; and representation of demographic groups in training and evaluation data is a baseline requirement, not an optional enhancement. These are non-negotiables, not recommendations.

Question 12

What Best Practices does GRACE define for ethical AI?

Accepted Answer

GRACE's Best Practices are actionable, implementable instructions organised across seven themes: Transparency (documenting societal risks explicitly and in advance), Accountability (using funding conditions as a lever for ethical review), Fairness (auditing AI systems for differential performance across demographic groups), Human Oversight (ensuring human control at all levels of automation), Stakeholder Inclusion (involving affected communities in research and product design), Data Governance (not assuming public data implies consent for research use), and Dual Use and Malicious Deployment (naming dual-use risks and responding to them concretely). Every best practice is directly traceable to a submitted input document.

Question 13

What Domain-Specific Considerations and Governance guidance does GRACE provide?

Accepted Answer

GRACE's Domain-Specific Considerations section records sector-specific guidance where sufficient consensus has emerged from submitted inputs. The current edition covers AI research and academic institutions — addressing funding conditions for ethics review, the structural distinction between ethics review and IRB review, professional society obligations, ethics education across computing curricula, and documentation of ethical issues for field-wide learning. The Governance and Accountability section accompanies this with design principles for effective ethics review bodies, including linking review to resource allocation, composing panels with disciplinary breadth, prioritising iteration over rejection, establishing transparent review criteria, extending review over project lifecycles, and building accountability mechanisms for unforeseen harms that emerge after approval.

GRACE — Global Repository of AI Consensus at Edyant

Principles, Policies, Guidelines, Guardrails and Best Practices for Ethical AI