The Governance Gap: Designing Global Standards for AI Advisory Boards

Abstract

The panel explored how AI advisory bodies can be shaped to deliver effective, transparent, and globally relevant governance. Sudhir set the stage by describing the uneven AI landscape—vast disparities in model size, company resources, and regulatory philosophies—from strong state‑led rules to self‑regulation. Julie explained the structure, powers, and early successes of Meta’s Oversight Board, emphasizing independence, binding decisions, and a human‑rights‑based framework. Saurabh described Sarvam AI’s practical experience navigating India’s nascent AI policy, highlighting the importance of continual regulator‑industry dialogue, data‑quality safeguards, and hybrid regulatory models that blend statutory baselines with company‑level self‑governance. The discussion then turned to institutional design, the need for “nimble” regulatory mechanisms, and how standards, certifications, and transparent reporting can bridge gaps for firms of all sizes. The audience raised questions about funding sustainability, liability regimes for AI‑generated content, enforcement of board recommendations, and the role of international bodies such as UNESCO, ITU, and the UN. Throughout, the panel underscored that robust, independent advisory boards—backed by clear accountability, human‑rights lenses, and multi‑stakeholder engagement—are essential to close the governance gap.

Detailed Summary

Moderator (Sudhir Krishnaswamy)

• Framed the session as an exploration of regulatory responses to a “very uneven” AI ecosystem.
• Key dimensions of unevenness
  1. Geographic divergence – a pronounced North‑South gap in large‑language‑model (LLM) development.
  2. Company size spectrum – from tech giants to tiny start‑ups, each with differing capacities for governance.
  3. Regulatory philosophy divergence – some stakeholders demand strong, government‑led legislation; others argue for self‑regulation that is independent, binding, and transparent.
• Stressed that AI is at a technological‑and‑political inflection point; the panel would focus primarily on institutional and regulatory design rather than purely technical mitigations.

Core Questions Posed by the Moderator

1. What should a minimum‑effective AI advisory board look like for organizations that care about safety, innovation, and profitability?
2. How can self‑regulation be credible (independent, binding, enforceable)?
3. What regulatory models (state‑led vs. industry‑led) can best accommodate the heterogeneous AI market?

2. Julie Owono – The Oversight Board as a Working Model

2.1. Origin & Scope

• The Oversight Board was created in 2020 by Meta (formerly Facebook) as an independent body that reviews content‑moderation decisions.
• Mandate – make binding decisions on whether specific pieces of content should be removed or kept, and issue policy recommendations to Meta.
• Composition – 22 members spread across five continents, representing 21 professional backgrounds (law professors, civil‑society activists, journalists, former editors ‑ e.g., Jakarta Post, The Guardian).

2.2. Governance Features

Feature	How it Works	Why It Matters
Independence	Members have fixed three‑year terms; cannot be fired by Meta’s leadership.	Shields decisions from commercial pressure.
Binding Decisions	When the Board rules that a piece of content must be removed (or reinstated), Meta is legally obliged to comply.	Guarantees real impact, not merely advisory “soft‑power”.
Policy Recommendations (Non‑binding)	Board may suggest broader policy changes; Meta must publicly explain any decision not to adopt them.	Enhances transparency and accountability.
Human‑Rights‑Based Framework	Decisions are filtered through rights such as free expression, safety, privacy, and the right to know.	Aligns AI governance with internationally recognised norms.
Transparent Reporting	Quarterly public reports on implementation status of recommendations (e.g., “75 % of recommendations applied”).	Enables external scrutiny and trust‑building.

2.3. Success Metrics & Impact

• Binding decisions have become a standard for Meta’s content moderation.
• Policy uptake: Approximately 75 % of the Board’s recommendations have been implemented, improving user safety and platform transparency.
• The Board’s human‑rights lens has been crucial in tackling AI‑generated content, including early deep‑fake cases (e.g., a falsified video of a former U.S. president).

2.4. Lessons for AI Advisory Boards

• Independence and binding authority are essential; advisory boards that are merely “consultative” risk irrelevance.
• Diverse, global membership ensures many perspectives (legal, civil‑society, journalistic) are considered.
• Clear reporting cycles and public accountability build credibility with both industry and the public.

3. Saurabh Karn – Navigating Regulation in India

3.1. Sarvam AI’s Governance Approach

• Proactive regulator engagement – maintains regular dialogue with Indian government bodies to anticipate requirements.
• Data‑quality focus – heavy investment in curated, high‑quality training data to reduce harmful outputs.
• Safety guardrails – employs red‑team testing, “dog‑fooding” (internal testing), and continuous monitoring before product release.

3.2. The Indian AI Policy Landscape

• India AI Guidelines (2023) favor innovation‑friendly regulation: promote growth while mandating accountability at company, industry, and national levels.
• Regulatory instruments – a blend of laws, guidelines, and self‑regulation. Recent “deep‑fake notification” order illustrates a move toward binding, targeted rules.

3.3. Anticipated Evolution

• Hybrid model: statutory baselines (liability, privacy) plus company‑level self‑governance.
• Industry standards & certifications (similar to ISO/ISO‑27001) will likely become market differentiators.
• Nimble regulatory cycles: regulators, industry, and civil society will need rapid feedback loops to keep pace with fast‑moving AI technology.

3.4. Challenges Highlighted

• Scale of diffusion – AI models are spreading faster than social‑media platforms did; global reach introduces cross‑jurisdictional complexities.
• Complex value chain accountability – responsibility must be apportioned across model developers, hardware providers, system integrators, and end‑users.
• Emerging agent‑based AI – delegated AI agents raise novel liability questions (who is responsible when an autonomous agent transacts on a user’s behalf?).

4. Institutional Design – Finding the “Middle Path”

4.1. Julie’s “Middle‑Ground” Argument

• Past lessons from social‑media: rapid innovation followed by unforeseen harms (e.g., hate speech, deep‑fakes).
• The Oversight Board demonstrates a proactive, independent model that can precede regulator action (e.g., early deep‑fake labeling).
• Hybrid solution: combine binding decisions (as with the Board) with policy recommendation mechanisms that invite broader stakeholder input.

4.2. Saurabh’s View on Governance Architecture

• Multi‑layered accountability: statutory rules set minimum safety baselines, while company‑run advisory boards handle day‑to‑day risk mitigation.
• Standards & certifications (e.g., safety benchmarks, ISO‑type audits) will serve as industry‑wide signals for compliance.
• Tech‑legal convergence: future regulation will need “techno‑legal” designs that embed technical safeguards (benchmarks, red‑team results) directly into legal obligations.

4.3. Open Questions & Points of Contention

Issue	Position(s)
Funding sustainability of independent boards	Julie notes current Reliance on a Meta‑managed trust; suggests exploring shared industry governance infrastructures to support smaller players.
Enforcement of advisory recommendations	Board’s binding decisions are enforceable; recommendations are not but must be publicly responded to.
International standard‑setting	Julie points to UNESCO’s 2020 AI ethics recommendation and UN working groups; both agree that global bodies have a role in shaping normative frameworks.
Liability for AI‑generated content	Sudhir/Julie acknowledge that existing product‑liability regimes can apply, but agent‑mediated actions create novel gaps that need further legal development.
Speed of regulatory response	Consensus that rapid, iterative cycles (stimulus‑response loops) are essential; over‑reliance on slow statutory processes is a risk.

5. Audience Q&A – Highlights

Questioner	Core Question	Key Points from Panel
Jyoti Pandey (Georgia Tech)	Funding model of the Oversight Board and its future sustainability; liability frameworks for AI‑generated content.	• Funding is held in a trust; sustainability being examined via industry‑wide shared governance. • Liability presently fits product‑liability but agent‑mediated transactions demand new legal concepts.
Caroline (unidentified)	Role of human‑rights in regulatory reform; internal governance for AI.	• Human‑rights‑based framework is central; it drives transparent, accountable decision‑making.
Vakasha (government liaison)	How enforcement of internal AI policies works amid competing business interests.	• Constant dialogue between advisory board and company leadership is essential; public reporting forces accountability.
Anonymous (ISO‑42001 auditor)	Emerging platforms (e.g., “Moldbook”) – how governance will adapt.	• New platforms will be subject to same standards; proactive benchmark/standard creation is needed to ensure compliance.
SD Saksana (telecom veteran)	Frequency of rejected board recommendations; possibility of delegating standard‑building to UN/ITU.	• Binding decisions are enforceable; recommendations are publicly addressed. • International bodies (UNESCO, UN Working Group) already drafting AI ethics frameworks; they can complement national efforts.
Unidentified participant	Complexity of large organisations and role of independent oversight.	• External independent bodies facilitate cross‑functional communication and prevent siloed decision‑making.

Notable Anecdotes

• Deep‑fake label experiment (2019‑2022): Oversight Board prompted Meta to add an AI‑generated‑content label, which was later adopted by billions of users.
• GitHub “agent‑contributed code” episode: An AI attempted to submit code; human reviewers rejected it, raising questions about agent‑generated user content and liability.

6. Concluding Remarks

• The panel agreed that no single model will fit every jurisdiction or company size; flexibility, independence, and enforceability are the non‑negotiable pillars of any effective AI advisory board.
• Collaboration among regulators, industry, civil‑society, and international bodies is essential to develop shared standards, certifications, and fast‑feedback loops.
• Future governance will likely evolve into a hybrid ecosystem where binding advisory boards co‑exist with self‑regulatory mechanisms and state‑level baselines, all anchored in a human‑rights‑centered approach.

Key Takeaways

• Independent, binding advisory boards (e.g., Meta’s Oversight Board) provide real enforcement and can pre‑empt regulatory gaps.
• Human‑rights‑based decision frameworks are crucial for aligning AI governance with global norms (free expression, safety, privacy, right to know).
• Diverse, global membership ensures that advisory boards capture multiple cultural, legal, and technical perspectives.
• Funding sustainability for independent boards may require industry‑wide pooled resources or shared governance infrastructures, especially to support smaller AI firms.
• Hybrid regulatory models—combining statutory baselines (India’s AI Guidelines) with company‑level self‑governance—appear most pragmatic for a heterogeneous AI market.
• Rapid stimulus–response cycles (continuous regulator‑industry dialogue) are essential; traditional, slow‑moving legislation will struggle to keep pace.
• Standards, certifications, and audits (analogous to ISO/ISO‑27001) can become market differentiators, incentivising firms to adopt safety best‑practices.
• Liability frameworks must evolve to address agent‑mediated actions and AI‑generated content, moving beyond classic product‑liability concepts.
• International bodies (UNESCO, UN Working Group, ITU) play a supportive role in shaping global AI ethics and standards, complementing national efforts.
• Transparent reporting (quarterly public updates) builds trust, holds companies accountable, and provides measurable evidence of advisory board impact.

---

See Also:

• navigating-the-ai-regulatory-landscape-a-cross-compliance-framework-for-safety-and-governance
• shaping-secure-ethical-and-accountable-ai-systems-for-a-shared-future
• supercharging-ai-adoption-in-the-global-south-opportunities-and-lessons-from-the-financial-sector
• ai-innovators-exchange-accelerating-innovation-through-startup-and-industry-synergy
• scaling-trusted-ai-for-8-billion
• enterprise-adoption-of-responsible-ai-challenges-frameworks-and-solutions
• policymakers-dialogue-on-ai-policy-evolution-and-the-rule-of-law
• leveraging-artificial-intelligence-in-public-audit-for-greater-transparency-and-accountability
• ai-for-fraud-prevention-and-financial-inclusion-in-bfsi
• founders-funders-the-india-ai-capital-ecosystem