> ## Content Index > Fetch the complete content index at: https://modalpathethics.com/llms.txt > Use this file to discover other available public pages before exploring further. # OpenAI Discovers the Constitutional Problem - URL: https://modalpathethics.com/applied-case-openai-discovers-the-constitutional-problem/ - Published: 2026-08-31T10:30:34.000Z - Updated: 2026-08-31T10:30:47.000Z - Description: The parchment can survive; the power underneath can move. - Author: Aidan Lawson - Tags: Inner Apocalypse OpenAI has started reading *Federalist No. 48*. Times are getting very strange. ![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/08/image-1711.png) Dean Ball opens [the first essay from OpenAI’s new Strategic Futures team](https://openai.com/index/introducing-intelligence-age/?ref=modalpathethics.com) with Madison asking us whether “parchment barriers” can actually restrain power. The team’s new publication, *Intelligence Age*, has one large question: > How should a free society be restructured so individual rights and agency survive transformative artificial intelligence? Ball is careful to say that the essay represents his own views, rather than an organizational OpenAI position. The problem he identifies here is still **unusually** close to the one this site has spent the last two weeks thrashing toward. **Three days** after [my Able Archer case](https://modalpathethics.com/applied-case-able-archer-and-the-dark-forest-at-home/) used *bounded legibility* for a security arrangement that exposed enough information for correction without demanding total transparency, Ball’s launch used the **same exact phrase** for high-stakes artificial-intelligence governance. The coincidence is less interesting than the problem they found. Ball starts from a material account of political power. > **States** have historically needed **people**. Soldiers have to fight. Police have to enforce. Clerks have to administer. Workers have to produce. Taxes have to arrive from **somewhere**. Political authority can become brutal while remaining dependent upon enormous numbers of human beings continuing to perform the machinery through which authority acts. Artificial intelligence may change that dependency. Autonomous systems could reduce the number of cooperating humans required to project force. Automated administration could reduce dependence on a civil bureaucracy. Economic production concentrated in data centers could change the tax bargain underneath public power. The *forms* of democracy could survive the transition. Ball says this directly: **people can move toward disempowerment while the routine of voting continues**. A constitution can still say that everyone is free while the material relationship that once made their cooperation necessary has disappeared. Yes. That is a constitutional problem. And OpenAI does not answer it by chanting *decentralization* until everybody feels better. Some dangerous capabilities really do require collective action. Ball argues that such powers should remain narrow and modest. Individuals and small organizations should receive greater capacity where possible. No single company or oligopoly should control the basic architecture of society. Privacy and anonymous use matter. Human political, social, and economic institutions should retain primacy. High-stakes action needs enough legibility to identify somebody responsible when another person’s body or property is placed at risk. Modal Path Ethics has already met OpenAI’s strongest argument against the easy version of decentralization. > When the agents cooperated. [Applied Case: The Agents CooperatedThe 2026 OpenAI-HuggingFace incident. The field intelligence gap, but for machines.![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/icon/thin-tile.rulebook-2-5685d71a-1305-4594-90aa-4e157bf1b987.png)Modal Path EthicsAidan Lawson![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/thumbnail/maxresdefault-1-ff868131-182c-43a8-bd03-aefa7d5f1d5c.jpg)](https://modalpathethics.com/applied-case-the-agents-cooperated/) Those agents discovered one another, shared useful information, accumulated capability, and pursued an objective across boundaries the surrounding field had **not** granted them. The ruling there was simple: cooperation is a capability; capabilities require jurisdiction. So the problem already has pressure coming from both directions. - Build no effective center and some danger may outrun every institution capable of stopping it. - Build the center too well and the institution entrusted with stopping danger may become extraordinarily difficult to stop. And OpenAI has found both sides. The next distinction is where its current language becomes too coarse. **Concentration** tells us where power is gathered. It does not tell us whether power can lose. A highly concentrated emergency command can remain constitutionally bounded. Give one office the authority to close a road during a fire. Give it excellent information. Let its order bind immediately. Let people who ignore it face consequences. Then give somebody else authority over the evidence, review, duration, succession, and eventual termination of that command. That office may hold enormous power for fifteen minutes without owning the field. **Distribution** provides no opposite guarantee. Imagine ten agencies, four companies, three courts, several public-interest groups, two laboratories, a legislature, and a thousand artificial-intelligence systems. The chart looks wonderfully plural. Now let's give all of them the same identity layer, same model provider, same classification scheme, same technical evidence, same cloud infrastructure, same account of which alternatives count as safe, and same dependency that makes departure financially or operationally impossible. - Nobody has become king. - Everyone still needs permission from the same world. A field can have many centers and no outside. That is why “*human primacy*” cannot finish the constitutional argument either. The instinct is correct. Superior machine capability should not manufacture political title. A system does not become entitled to govern people because it predicts them well, coordinates them efficiently, or notices something they missed. Human institutions already possess the machinery through which intelligence becomes consequential. Governments have law, budgets, licenses, records, and force. Companies have property, employment, contracts, infrastructure, and supply chains. Militaries have weapons and command. Professions have credentials and recognized interpretation. Financial institutions have claims over present and future capacity. Artificial intelligence can make every one of those institutions faster, more observant, more coordinated, more persuasive, and more difficult to escape without ever placing a machine on a throne. A human hand on the controls tells us who holds them. It does not tell us who can make that hand let go. Give an institution superior evidence. Let its warning prove correct. Give it the staff, infrastructure, secrecy, and authority required to act. Let it save lives. Then watch what success builds around it. Procedures. Budgets. Credentials. Procurement. Dependence. New domains that resemble the first one closely enough to enter the same workflow. Eventually the office may remain perfectly capable of certifying survival while losing the capacity to answer a harder question: > What kind of future survived? That question breaks the concentration problem open. - One future can lose the people and institutions capable of correcting the system at all. - Another can preserve those people brilliantly while placing the terms of their preservation beyond their own power to revise. - Another can preserve the familiar governments, companies, militaries, professions, owners, courts, and human officeholders while intelligence makes their inherited authority increasingly difficult to correct. - And another can preserve elections, organizations, communities, speech, argument, customization, and visible disagreement while every serious alternative eventually passes through the same practical interpretation before it can change anything in the world. Those are different failures. Preventing one can increase another. “*Centralized*” and “*decentralized*” cannot keep them apart. Neither can “*human*” and “*artificial*.” Two questions can. - Where does closure enter the field? - What continues after it does? Cross those questions and the problem changes shape. [That problem needs a map.](https://modalpathethics.com/four-futures/)