Open Global Investment as a Governance Model for AGI

Abstract

AGI development could proceed within one or more corporations in a context that encourages wide international shareholding, reduces expropriation risk, strengthens corporate governance, operates within a government-defined responsible-AI framework, and admits additional international agreements. The model is very imperfect, but offers advantages in inclusiveness, incentive compatibility and practicality over prominent alternatives — especially under short AGI timelines.

Working paper, version 1.15 (October 2025); first version July 2025.

Framing

• Governance models emphasise different objectives — democratic oversight, national security, international cooperation, legitimacy, economic efficiency, equitable benefit sharing, AI safety, free enterprise, wise stewardship, scientific advancement, operational security, responsible deployment. No structure can fully achieve all of them; the choice is between imperfect options.
• Transformative AI adds special challenges: existential catastrophe from misaligned or misused AI, and extreme concentrations of power.
• As an intelligence explosion approaches, governance structures come under great strain. It is therefore not enough that a structure would be satisfactory if adopted — it must be robust enough to maintain integrity throughout the process and incentive-compatible enough to be adopted in the first place and retain support.

The model

Variants

• OGI-1 vs OGI-N — a single lead AGI company versus several.
• US-OGI — companies domiciled in the United States. The analysis is US-centric but the model is geographically neutral.
• The status quo already roughly approximates OGI. Favouring the model therefore means both striving toward its ideal form and, perhaps more importantly, resisting proposals such as nationalisation that move further away from it.

Illustrative US-OGI-1 structure

• (a) A publicly traded “AGI Corp” domiciled in the US — either a standard corporation or a Delaware public benefit corporation.
• (b) Widely distributed ownership: individuals worldwide, sovereign wealth funds and other investment vehicles; any IPO structured for broad initial distribution.
• (c) Multiple share classes separating profit participation from voting rights — illustratively Class A (1 vote), Class B (10 votes), Class C (1,000 votes). Class A might be open worldwide; B and C restricted to citizens of countries agreeing to a responsible AI framework; C allocated as founder shares. A limiting case is a golden share, which if held by a government amounts to soft nationalisation.
• (d) Foreign governments and citizens — including strategic rivals — are encouraged to buy shares, perhaps up to a cap such as 20% per country.
• (e) Cooperating governments could help AGI Corp achieve a large lead through subsidies, tax breaks, regulatory waivers, and regulation impeding competitors (e.g. chip access restrictions).
• (f) AGI Corp could acquire and merge competitors, with antitrust enforcement waived, aided by a low cost of capital from its sanctioned status.
• (g) Charter provisions strengthening governance beyond typical Fortune 500 practice — more frequent board meetings, enhanced internal oversight resources.
• (h) A significant share held by independent organisations and not-for-profits with humanity-benefiting remits.
• (i) AGI Corp remains subject to the law of its domicile: the USG can block unsafe deployment, halt development pending safety and security standards, and vet technical personnel against espionage.
• (j) Legal and structural devices making expropriation by any government as difficult and costly as possible.
• (k) Periodic public commitments from government leaders respecting property rights and independence — no confiscatory taxes or nationalisation. Ideally embedded in law and treaty; informal pledges are better than nothing.
• (l) Data centres distributed across multiple countries and jurisdictions, security permitting, to further raise the cost of expropriation.

In OGI-N versions, (e) and (f) do not apply, leaving something closer to today’s status quo. The paper is agnostic between OGI-1 and OGI-N.

Regulatory context

• Ownership and control is only half the picture; the other half is the rules governments impose on the sector — consumer protection, political integrity, worker displacement, abuse prevention.
• In a US-OGI scenario the USG would bear primary oversight; other countries would regulate domestic use. Ideally states agree an international framework providing basic global safeguards within which national rules operate.
• The corporation itself can decline products it judges harmful to shareholders’ long-term interests or its mission; a PBC charter widens the scope for this, though ordinary boards already have considerable latitude in practice.

Motivations

1. Leverage entrenched property-rights norms, laws and institutions, which may be more robust and reliable than a novel ad hoc scheme concocted for AGI.
2. Distribute ownership and control widely, which:
   • gives powerful actors personal incentives to uphold property rights rather than push for expropriation that would unravel the framework protecting their own interests;
   • offers international rivals an alternative to resentment and hostility, and may at the margin reduce their willingness to race;
   • promotes a wider and more globally equitable distribution of benefits and influence.
3. Present a path with a realistic chance of implementation given current actors’ motivations, geopolitical constraints, and short timelines.

Roles and powers of actors

• AGI corporation(s) choose products, customers, prices and use constraints. Because products may affect shareholders far beyond direct profits, the corporation would plausibly weigh non-monetary shareholder interests more than a typical company. In OGI-N, competitive pressure reduces any one firm’s ability to forgo a harmful product — an argument for OGI-1, or at least a very small number of top firms.
• The host government can block developments and deployments, mandate anti-espionage procedures, and choose whether to assist via stock purchases, deregulation, subsidies or hindering competitors. In a fully realised OGI it also gives up options — committing not to nationalise, not to impose punitive taxes, and not to force foreign divestment.
• Other governments have six channels of influence: buying shares; regulating products in their jurisdiction; suing in US courts if a PBC fails its public benefit mission; diplomatic and economic pressure on the host; recourse under any multinational agreements; and, as a last resort, seizing corporate assets located on their territory. The paper also floats novel technical arrangements such as an n-of-m multisignature mechanism to remotely disable critical AI hardware under specified conditions.
• Citizens retain ordinary political and consumer channels, plus the option to buy shares — distributed unequally by wealth, though national governments, sovereign wealth funds, religious organisations, charities and NGOs could buy on behalf of poorer constituents.

Representativeness and fairness

• The assumption that a government-controlled initiative is more representative is questionable: the US has about 4.2% of world population and 26% of nominal GDP, so a nationalised project excludes 95.8% of humanity and 74% of the world economy — and that assumes it stays under effective democratic control, which is far from given.
• Rough Gini comparison: global wealth ≈ 0.89; global stock ownership plausibly 0.90–0.92; a US-nationalised programme (even assuming perfectly equal stakes among US citizens) ≈ 0.96. By this measure nationalisation is more unequal than OGI.
• The gap widens further when accounting for taxation of AGI-sector profits in both host and shareholder countries, and in OGI-N where monopoly profits are partly competed away.
• More inclusive arrangements are conceivable — a UN-run project — but may not be incentive-compatible for current power-holders and funders, and international organisations are often slow and may not be capable of running a globally competitive effort.
• Historical note offered as caution: congressional oversight of the Manhattan Project was extremely limited, and Vice President Truman did not learn of it until after Roosevelt’s death.

Objections addressed

Military applications and foreign competitors

• Even non-military-specific AGI could greatly aid military planning, logistics, drone and robot operation, rapid military R&D, and cyber and information operations; rapid economic growth alone could destabilise geostrategic positions.
• The host government may therefore restrict sales of top-tier capabilities abroad — and perhaps even to its own citizens on dual-use grounds — prompting rivals like China to pursue independent projects even if invested in AGI Corp.
• But the relevant comparison is to implementable alternatives, not an idealised harmonious world order. Under a Manhattan Project or privately held model, rivals’ incentives to pursue independent projects would be stronger; under OGI they at least share in the economic upside and have some say via voting rights. This may reduce the urgency or scale of rival efforts at the margin, and could conceivably tip the balance if racing were shown to pose extremely high existential risk.
• OGI is as amenable as other models to augmentation with arms control agreements and non-aggression pacts.

Private control and national security

• Precedent: most advanced US defence materiel comes from privately owned, publicly traded firms (Lockheed Martin, RTX, Boeing, Northrop Grumman), over which the USG exerts extensive control via monopsony power, security clearances, ITAR, DCMA embedding, the Defense Production Act, and CFIUS review with FOCI mitigation.
• Investors may reasonably fear de facto expropriation. Defence contractors mitigate this through lobbying, revolving-door hiring and geographically distributed factories — but AGI firms may be more vulnerable, since windfall intelligence-explosion profits and national-security centrality make them a tempting political target, and unlike defence contractors they probably won’t create vast numbers of jobs relative to market cap and may be blamed for automation-driven job losses.
• Invoking Chinese competition might fend off shutdown, but long-term autonomy requires broader elite buy-in — which is what wide shareholding provides.

Speed gap between technology and regulation

• In fast-takeoff scenarios, any model relying entirely on traditional regulatory processes will fail to keep pace; regulation is often the wrong instrument for unique, speculative or highly technical judgment calls.
• The analogy: a military campaign cannot be managed by issuing regulatory codes but requires field commanders exercising judgment — so outcomes will depend heavily on the values and competence of specific individuals in key positions.
• This implies OGI may require unusually close coupling between company and government: regular meetings with a dedicated technical task force, embedded government representatives, a continuous oversight board, or at the extreme a public-private partnership — with extra safeguards to reassure shareholders.

Transition to post-AGI governance

• OGI is intended for the intermediate stage between now and full-fledged superintelligence. Beyond that, fundamental changes may be necessary, but by then the situation should be clearer and decision-makers may have superintelligent advisors.
• It remains relevant longer-term insofar as it helps reach that point with minimal negative-sum conflict and shapes the initial conditions for whatever follows.

Discussion

• OGI is an “ideal type” in the Weberian sense — not necessarily first-best — which reality may more or less closely approximate. As of mid-2025 the status quo already delivers some of its benefits: Alphabet and Meta are publicly traded, as are the dominant AI chip designer, leading foundries, equipment makers and largest data centre providers; even unfriendly countries can gain diluted exposure via SoftBank, BlackRock and similar.
• The “astronomical bonanza” counterargument: if outcomes are vast enough, even highly diluted exposure suffices for any resource-satiable priority — “whether you end up owning 1,000 galaxies or a paltry 10 solar systems.” One slightly generous holder of 1,000 galaxies spending 0.1% philanthropically could give every existing person 10 solar systems.
• Why that isn’t decisive: the scenario may not hold under the simulation hypothesis or with other claimants to the cosmic endowment; behaviour of states, charities and philanthropists may change after such transformations; many actors have non-satiable positional preferences; and even if their long-term preferences would be satisfied, the scenario may not feel realistic enough during the run-up to dissuade desperate moves — whereas a legible equity position offers more reassurance.
• The model’s value lies partly in arguing against moves away from it: nationalisation on inclusivity grounds excludes 95.8% of humanity; nationalisation to avoid a safety race-to-the-bottom ignores that OGI lets foreign elites invest rather than compete or sabotage; concerns about coups and destabilisation favour well-established property and corporate-governance norms integrated with civilian society; and new international organisations should be compared on political feasibility, time to fruition, security, funding and organisational efficiency.
• The UN’s 2024 Governing AI for Humanity report proposes modest steps (a scientific panel, capacity building, coordination) illustrating how far international proposals remain from governing AGI development. An IAEA-style agency could take years or decades and would still face enforcement, agility and competence challenges. OGI can be implemented immediately while remaining compatible with international coordination focused on standard-setting, monitoring or enforcement rather than ownership.

Appendix 4 — comparisons with other models

vs “Manhattan project for AGI”

• More agreeable to incumbents (company leadership, personnel, investors); wide investability increases incentive compatibility among US and international elites.
• Wider and more equitable distribution of benefits and influence.
• Obviates massive government funding.
• May reduce extreme power concentration via a dual veto structure between corporation and government, and by embedding the project in civil society rather than the security state.
• Mitigates negative-sum racing by letting many countries and elites participate.
• Offers a range of government-involvement options, from status quo through to soft nationalisation.
• Compatible both with cooperative international frameworks and with unilateral US influence via the semiconductor supply chain.

vs “CERN for AGI”

• Inherits the Manhattan model’s downsides — less incentive-compatible for incumbents, needs massive government funding, slow to set up, bespoke and less-vetted organisational construction.
• Adds its own: formidable information security challenges with globally drawn staff under diplomatic protections; unclear whether it could match corporate development speed.
• Its advantages: greater global equity and potentially greater acceptability to great powers.
• Critically, its mere existence would not eliminate competing corporate or national projects. Achieving a de facto global monopoly would require a binding international agreement between all capable actors — and if circumstances made that feasible, a similar arrangement might also be available for OGI (though states might be more willing to subordinate their efforts to a CERN than to a US AGI Corp).

vs “Intelsat for AGI”

• Intelsat was an intergovernmental consortium and treaty organisation where member countries were shareholders with voting power proportional to investment and usage — moderated by a two-tier board (one chamber with one vote per country), regional quotas, supermajority requirements, and a gradual transition away from initial US dominance.
• Closer to OGI than CERN is, in having a genuine business component with expected and realised financial returns.
• Key difference: OGI allows participation by private individuals and corporations, which gives participating countries greater confidence against asset seizure, since personally invested economic and political elites have incentives to protect the project’s autonomy — analogous to multinationals bringing prominent local business families on board as co-investors in weak-rule-of-law jurisdictions.
• Like the others, Intelsat did not prevent rivals: the Soviet Union built Intersputnik, and several countries deployed national or regional alternatives.