Artificial General Intelligence - By Sebastián Muller
For the purpose of this assignment, I will resort to Black’s “decentred” understanding of regulation, recognizing that it may originate from governments as well as social actors. Therefore, my chosen institutions for Artificial General Intelligence (AGI) governance are the following:
1) US executive branch (government): its authority stems from the President's constitutional power to supervise and direct the executive branch. The main regulatory instruments used are policy documents as well as executive orders. Examples of these are the National Policy Framework for AI from March 20, 2026 (1), as well as the executive order titled "Ensuring a National Policy Framework for Artificial Intelligence" issued on December 2025 (2). The self-declared purpose of the latter is to ensure the United States’ leadership in Artificial Intelligence, promoting the national and economic security and dominance. As for weaknesses, while National Policies issued by the President are considered lawful if used to steer the executive branch, the same cannot be said regarding certain executive orders. For instance, the 2025 order has been deemed unlawful by many critics, as it could override powers from the legislative branch (3).
2) AI Office (public institution): its authority stems from the Commission Decision of 24 January 2024 (4), as well as the legal powers granted to it through the AI Act. The main governance instruments used by the Office vary from soft guidance documents (such as guidelines) to a sanction-backed enforcement. Its declared purpose is to protect the health, safety and fundamental rights of people within the AI environment. Finally, its main weakness could be its lack of capacity, mainly due to the complex nature of this technology. Going back to Black’s framework, government institutions may observe AI systems from the distorted lens of their perceptual apparatus; thus, it is possible that they lack the industry insight to properly regulate and conduct social behaviors. Additionally, the Office currently lacks powers regarding this specific technology, as it is not being developed within the EU territory.
3) AI Customer base (civil society): both Lobel and DeNardis/Raymond recognize that individuals organized within the civil society are institutional actors. In this case, an organized group of consumers sued the main AI firms, accusing them of slowing down the development of their models and diminishing the value of their subscriptions (5) (hence, using the court system and public pressure as their main instrument). Their authority originates in the contractual relation between AI providers and its customer base. Their main interest is to ensure product and service quality. Finally, their main weakness is their lack of accountability within a governance ecosystem, as the customer base does not respond to any other stakeholder than itself.
4) AI firms (incumbents): their authority is mainly factual (as they are the only ones with key know-how to develop this technology), although it is backed by a favorable US National Policy. Firms resort to self-regulation and public pressure as key instruments to obtain their goals, although recently they have been asking for government regulations due to safety concerns (6). While they claim to value product safety, arguably their main interest is to prevent other actors to enter the market and, therefore, increase their own profits. Two weaknesses can be mentioned: first, firms usually operate under their own undisclosed standards, lacking proper accountability; and second, now that they have attempted to standardize their practices towards a slow-down, antitrust issues have been flagged (7).