Shallow Review of AI Governance: Mapping the Technical–Policy Implementation Gap

Project summary

Across AI governance, there is a growing menu of regulatory proposals, yet most struggle at the point of implementation. Technical alignment approaches often assume enforcement capacities that policymakers don’t currently have, while policy proposals frequently fail to account for core technical failure modes such as misaligned generalization or speculative control and enforcement assumptions.

I propose a 4-month empirical “Shallow Review” of the AI governance macrostrategy landscape focused on operational enforceability. This project extends prior Shallow Reviews by explicitly stress-testing alignment and control mechanisms against real enforceability constraints, surfacing operational cruxes at the intersection of technical safety and governance implementation.

The aim is to close the inferential gap between technical AI safety (what alignment requires) and policy (what can be executed across jurisdictions and institutions). Rather than adding more theory, this project evaluates implementation: What could actually be enforced? Under what institutional capacity? And where are the bottlenecks that block practical governance?

This work contributes directly to open needs around macrostrategy, coordination, and “connector-style” translation between technical and policy communities, using lightweight expert elicitation to identify where consensus is possible, and where disagreement represents a true strategic crux rather than simple terminological confusion.

This is a 4-month shallow review with publishable artifacts:

operational scoring of governance pathways
expert-informed cruxes
enforceability rubrics
a translator’s guide for policymakers and technical researchers

With the intention of creating a public good, all outputs will be open-licensed, reusable, and structured so future researchers can extend or fork the taxonomy.

What are this project's goals? How will you achieve them?

Goals:

Map the Landscape: Build a taxonomy of 30–50 concrete governance mechanisms (e.g., licensing, evaluation requirements, hardware oversight, liability schemes, international standards), grading them on technical feasibility vs. operational viability.
Map Macrostrategy Cruxes: Run a lightweight Delphi process (two survey rounds, 15–20 experts) to identify where experts converge or sharply disagree on governance cruxes (e.g., enforceability of compute caps, viability of licensing), and where alignment/technical and policy views diverge.
Score Viability: Simple comparative rubric on feasibility and enforceability
Create a Shared Vocabulary: Produce a 2–3 page "Translator’s Guide" that makes technical safety assumptions legible to policymakers, and policy constraints legible to technical researchers.

Methodology (4-Month Timeline):

Month 1: Ingestion & Taxonomy. Review ~50–70 key documents (GovAI, CSET, AI labs’ safety/governance reports and similar). Public "Draft Governance Taxonomy" for community feedback.
Month 2: Delphi Round 1 (Divergence). Lightweight expert sampling (20–30 mins) to identify disagreements, constraints, and failure points. Experts from GovAI, CSET, AI labs’ governance teams, and similar to establish a baseline of disagreement on operational bottlenecks.
Month 3: Delphi Round 2 (Convergence). Feedback of Round 1 data (medians and anonymous arguments) is returned to experts. This iterative step is critical for distinguishing true ideological "Cruxes" from simple misunderstandings. Views based on anonymized summaries. Success is signal extraction, not academic-level consensus
Month 4: Synthesis & Publication. Final synthesis of consensus data and publication "Shallow Review" (8–15 pages), browsable taxonomy (Notion/wiki), and Translator’s Guide, all public on Alignment Forum and Manifund.

How will this funding be used?

I am requesting $23,500 to fund a 4-month deep dive to prioritize researcher throughput and high-quality expert engagement.

Research Stipend: $20,000

Expert Honoraria: $2,500

Compute, Hosting, & Tools: $1,000

Minimum Funding Scenario:

With only $8,000, I will deliver an MVP. Still delivers a high-value field map; Delphi deferred unless additional funds arrive.

literature review
draft taxonomy
initial operational scoring
public publication

The full funding enables the Delphi component and much stronger signal extraction.

Who is on your team? What's your track record on similar projects?

Advisor: Dr. Chinasa T. Okolo:

Chinasa T. Okolo, Ph.D., is the Founder of Technecultura, a Policy Specialist at the United Nations Office for Digital and Emerging Technologies (ODET), and a recent Computer Science Ph.D. graduate from Cornell University. Her research focuses on AI governance and safety for the Global Majority, datafication and algorithmic marginalization, and the geopolitical impacts of AI. Dr. Okolo has been recognized as one of the world’s most influential people in AI by TIME, honored in the inaugural Forbes 30 Under 30 AI list, and advises numerous multilateral institutions, national governments, corporations, and nonprofits. She is a former Fellow at the Brookings Institution and has worked in research-based roles at Apple and Microsoft. Her research has been covered widely in media outlets and published at top-tier venues in human-computer interaction and sociotechnical computing.

Researcher: Anthony Ware

I am an operational strategist transitioning to full-time AI Macrostrategy after 15+ years in startup and organizational strategy, coordination environments, and decentralized governance with clients across multiple countries.

My contribution is implementation realism: identifying where proposals fail when they meet institutions, constraints, and incentives.

I operate as a “translator” between technical and policy communities, reducing inferential distance and building legibility, precisely the connector role the field has identified as underdeveloped outside academia.

While I haven't done this exact study before, I understand the process. I did release the Founder Mental Wealth Report, a research and synthesis project involving a sensitive population, expert interviews, and turning a complex topic into public output. This involved mixed-methods analysis of previous research studies and reports, surveys and interviews of over 200 founders in the US and UK involving sensitive qualitative and qualitative data collection, and organizing the data for public release in a way that was actionable and accessible.

Designing environments that protect identities and still surface signals is familiar territory for me. My experience here comes from years of working with founders, ERG leaders, and other groups where privacy, reputational concerns, and psychological safety were real barriers to candid participation. In previous research, anonymity and minimal time commitment helped to increase participation, especially from people who were cautious or didn’t want to attach their name to a particular stance.

My expectation is similar for this project. I am deliberately asking for a signal around disagreement, not deep or sensitive material. That is why the design flow is taxonomy first, then Delphi to reduce risk and make participation feel like a contribution to a larger conversation, not personal. Also, I realize and have the experience of outreach being a numbers game in terms of success...meaning to get 10 participants, I would potentially need to contact 30+ people. The methodology of the project is part of my strategy for navigating secrecy and reputational caution, something past experience has taught me.

The "Translator" Advantage: As a non-technical macrostrategist, I convert high-context technical safety concepts into low-context policy frameworks. This directly addresses the "Connector" gap identified in the ecosystem, ensuring that technical insights are actually legible to the stakeholders who need to enforce them.

Track Record: Empirical Research & Synthesis: Unlike many independent researchers who struggle to ship final products, I have a track record of executing and publishing complex projects:

Founder Mental Wealth Report (2019): I designed, executed, and published an empirical study on 179 founders across the UK and USA. This demonstrated my ability to gather sensitive qualitative data, synthesize it into a coherent framework, and publish a legible artifact for the community.
Technical Synthesis: Authored Flexibility Not Required, a 400-page training manual that deconstructed complex physiological concepts into actionable protocols. This mirrors the exact work required for the "Translator’s Guide" deliverable in this proposal.

Experience: Operational Security & High-Stakes Coordination

Dual-Use Tech Transfer: As Co-Founder of Trias Global, I navigated the "Valley of Death" in technology transfer, establishing commercialization protocols with the NSA and Purdue’s CERIAS (Center for Education and Research in Information Assurance and Security). While the venture was early-stage, the experience gave me native fluency in the bureaucratic friction that stifles hardware governance and security control.
Polycentric Governance: As Former Lead Steward of the Public Goods Working Group at ENS DAO, I managed budget allocation in a decentralized, non-state environment. This provides a unique lens on "International Governance" where no single sovereign exists. My goal here is to work as a translator and implementation auditor: aligning technical safety constraints with policy implementation pathways.

The "Non-Technical" Advantage

Global Operational Fluency: Having worked with professionals across the Australia, US, UK, and EU, I possess the cross-cultural "code-switching" ability necessary to harmonize international governance frameworks.
Operational Stress-Testing: My background allows me to "Red Team" policy proposals for operational viability (e.g., "Can this actually be enforced?"), ensuring the final "Crux" map is grounded in reality, not just theory.
Native Policy Fluency: While technical experts may struggle to decouple their insights from mathematical jargon, I function as a 'Translator.' I can distill complex alignment failure modes into the operational language required for governance implementation, reducing the inferential distance for policymakers."
Epistemic Neutrality: I am not committed to a specific technical agenda (e.g., Mechanistic Interpretability vs. Evals). This allows me to audit the governance landscape as a neutral observer, aggregating disparate theories of change without the 'agenda bias' common in technical grant applications.

What are the most likely causes and outcomes if this project fails?

Cause: Low participation in the Delphi study (Experts are busy).

Outcome: The project pivots to the MVP. While this loses the dynamic consensus element, it still provides a high-value "Map" of the field’s current state.

Cause: A "Null Result" (e.g., finding no consensus or viable governance paths).

Outcome: I commit to publishing a "Negative Result" Post-Mortem. A rigorous map showing exactly where the pathways are blocked is valuable information for the ecosystem and future grantmakers. Even null results help redirect resources and highlight blocked governance pathways.