To break the monopoly that shareholder interests hold over AGI, we must move beyond voluntary guidelines. Relying on companies to self-regulate is a losing game. Instead, we need to bake public interests into the legal foundations of these entities.
One approach involves creating multi-stakeholder oversight boards. These wouldn't just consist of engineers and executives. They would include civil society representatives, ethicists, and labor advocates with actual veto power over high-risk deployments. If a model poses a systemic risk to democratic stability or labor markets, these boards must have the authority to halt development.
We also need to redefine corporate law. Currently, directors often feel legally bound to maximize stock prices. Legislators could introduce 'public interest' mandates. This would legally require boards to weigh societal impacts alongside profit margins. If they ignore these risks, they would face personal liability, not just corporate fines.
Finally, transparency cannot be an afterthought. We need mandatory, standardized audits of compute usage and training datasets. Open-source development also serves as a check. By democratizing access to the underlying tools, we prevent a few private boards from gatekeeping the future of intelligence. We must treat AGI as a public utility rather than a private toy.
