Leaders and researchers in the AI industry do not compete for authority by saying they want power. They compete by invoking technical, ethical, and civilizational languages that frame their claims as fidelity to safe AGI development, loyalty to accelerating human progress, or responsibility for shaping the future of intelligence inside a hyper-competitive, post-ChatGPT global race. This is the core insight of David Pinsof’s Alliance Theory. They recruit allies, define legitimacy, and justify control over institutions, frontier labs, compute clusters, regulatory influence, talent pipelines, funding rounds, and the invisible networks of model releases and safety benchmarks. The key language is not only technical. It is also practical and social. Being summoned. Building beneficial AGI. Prioritizing alignment. Accelerating responsibly. Democratizing intelligence. These phrases do not merely describe practice. They define jurisdiction. They determine who gets to say what kind of AI future the industry can sustain, how demanding the safeguards should be, and which forms of balancing still count as faithful.
Before the analysis proceeds, the framework needs a limit acknowledged. Alliance Theory, applied without restraint, becomes a closed system. When every position gets decoded as a power move, the analysis loses precision. The engineer who stays up until 3 a.m. scaling the next model is not primarily executing a coalition maneuver. He is trying to maintain a form of frontier research life he genuinely values. The core values, alignment, capability, beneficial intelligence, responsible stewardship, carry real internal logic and authority for those inside. Alliance Theory names something real about how institutional authority functions in the AI industry. It is not the whole picture.
Ernest Becker argues in The Denial of Death that human beings are unique among animals in their awareness of their own mortality, and that most of human culture, religion, and social life organizes itself to manage the terror that awareness produces. We construct hero systems, cultural frameworks that promise symbolic immortality, that tell us our lives participate in something larger and more permanent than our individual bodies. To be a faithful member of a hero system is to transcend death symbolically. To lose one’s hero system is to be thrown back against the terror it was built to contain.
The AI industry is a hero system of unusual density, and it operates at a scale of existential stakes that no other institution in this series can match. Every other domain examined here, philosophy, Goldman Sachs, McKinsey, the PMC, the spirituality business, offers participants symbolic participation in something larger than themselves. The AI industry offers something more literal. Its practitioners genuinely believe they are building the last technology humanity will need to invent, the one that invents everything else. The civilizational silence that follows a failed or misaligned transition to AGI functions in this community as the collective version of Becker’s individual death terror. The jurisdictional war is therefore not merely a struggle over institutional control. It is a struggle over who acts as gatekeeper for humanity’s transition into a post-biological era. The stakes feel existential because, in this case more than any other, they might actually be.
Every model release carefully evaluated for safety, every compute cluster built with long-term impact in mind, every refusal to chase the latest hype cycle: these are not merely professional obligations. They are acts of fidelity to a post-2012 heritage that has sustained frontier AI development through conditions far worse than the current era of geopolitical competition and trillion-dollar valuations. That is a hero system. It promises that an individual life, lived seriously within this framework, participates in something that neither death nor the surrounding culture of quarterly earnings can fully dissolve.
Scaling laws function as this industry’s dogma in a way that has no equivalent in any other domain examined here. They provide a quasi-religious assurance that labor, in the form of compute and data, will be rewarded with a proportional increase in intelligence. When a model fails to show expected emergent properties, it is not merely a technical setback. It is a crisis of faith. The safety coalition treats scaling laws as a warning of an approaching force that must be constrained through alignment work. The accelerationist coalition treats the same laws as a moral imperative, where withholding compute becomes ethically equivalent to withholding a cure for a fatal disease. Both sides read the same empirical pattern and reach opposite normative conclusions because the normative conclusion was never really derived from the empirical pattern. It was present before the analysis began.
Iddo Tavory’s concept of summons, developed in Summoned: Identification and Religious Life in a Jewish Neighborhood, adds the theoretical layer that explains how the hero system reproduces itself. The AI industry is not simply a place where researchers happen to work near one another. It is a network in which people are repeatedly called into being as builders of the future through institutions, interactions, conferences, safety workshops, compute allocations, and ordinary Slack-side recognitions. The industry’s thickness is the product of repeated summons into frontier intelligence being. To belong here is to be hailed, continuously and from multiple directions, as a particular kind of pioneer.
Through Becker’s lens, those summons are not merely social. They are the hero system doing its maintenance work. Each summons interrupts private drift, which in Becker’s terms means each summons interrupts the moment when the individual is thrown back toward unmanaged anxiety about irrelevance or civilizational failure. That is why defection carries such disproportionate social weight. The researcher who questions the rush toward larger models or who begins softening safety protocols to ship faster when his circle holds firm is not merely making a technical adjustment. He is, in the community’s felt logic, weakening the collective structure through which everyone present manages the terror that humanity’s future was built to contain. Authority in this industry is enforced through epistemic exile. A researcher who moves from a frontier lab to a commercial product team is often spoken of as having left the priesthood. The loss of seriousness is the primary punishment, and it is administered not through formal sanction but through the withdrawal of the summons itself.
Three master domains organize the struggle over institutional authority. The first is moral authority over what counts as responsible AI development. The second is the organizational structure of frontier labs, compute infrastructure, talent acquisition, and regulatory influence. The third is the everyday network through which AI distinction gets reproduced in model releases, safety evaluations, conferences, and the mundane problem of navigating the field without becoming geopolitically or commercially porous.
The hardline-safety coalition, concentrated around Anthropic under Dario Amodei and Google DeepMind under Demis Hassabis, uses the language of rigorous alignment and separation from reckless acceleration. Its claim is that the industry’s value lies precisely in its capacity to sustain careful, safety-first development against the pressures of geopolitical competition and market hype. In Becker’s terms, this coalition defends the integrity of the hero system against the accommodations that slowly evacuate it. Every softening of the summons is experienced not merely as a technical compromise but as a threat to the structure through which the community manages its civilizational stakes.
Against this stands the accelerationist coalition, strongest among those pushing frontier capabilities including xAI under Elon Musk and elements within OpenAI under Sam Altman, using the language of responsible speed, workable deployment, and competitive necessity. Their claim is not that safety should be abandoned. It is that AI development cannot be governed as though it were still a pure academic exercise. Once one side defines the industry’s purpose as sustaining maximal safety rigor, speed begins to look like recklessness. Once the other side defines the industry’s purpose as winning the global race under actual competitive conditions, maximal caution begins to look like strategic abdication masquerading as virtue. Neither side says it is fighting over compute contracts, talent pipelines, regulatory capture, or the trillion-dollar valuations that flow to whoever establishes the dominant narrative. Each says it is protecting the true future of intelligence.
The gap between stated values and operational reality is visible in the capital flows. In early 2026, OpenAI reportedly closed a record $110 billion round while xAI secured $20 billion, both focused primarily on infrastructure, chips, data centers, and power, to push the scaling frontier. Anthropic, the safety coalition’s standard-bearer, raised $30 billion. Philanthropic and independent funding for global coordination, treaty frameworks, and international governance sits at less than $5 million annually, a ratio of roughly twenty to one against technical safety research and far larger against frontier capability investment. The signal layer of the industry speaks constantly about alignment, stewardship, and beneficial intelligence. The cue layer speaks in compute allocations and funding rounds. Participants learn to read cues.
The open-versus-closed debate adds a further jurisdictional layer that has no equivalent in the other domains. Open-source advocates frame their work as democratizing intelligence, a moral vocabulary that recruits the public as an ally against regulatory capture by the closed labs. Closed-model labs frame their secrecy as responsible stewardship, arguing that transparency in the face of existential risk is a form of negligence. This creates a logic where even the act of hiding one’s work becomes a marker of higher fidelity to the mission. Both positions are genuine, and both are also coalition technologies. The open-source frame recruits a broad public coalition and positions the closed labs as an illegitimate guild. The closed-model frame recruits regulators and serious researchers and positions open-source advocates as naive or reckless. Neither side acknowledges that its moral vocabulary also happens to serve its competitive position.
The shift toward Sovereign AI introduces a further fracture that the original hero system was not built to handle. Nations building their own localized frontier models are importing the language of the global AI hero system while deploying it in service of national interest. The original hero system was global and implicitly universalist. The new system is Westphalian. Leaders must now balance their summons as global stewards of humanity’s future with their role as strategic assets for specific states. The accelerationist coalition frequently uses the threat of adversarial AGI development by China to justify faster domestic development, effectively merging Becker’s civilizational hero system with the older and more primal hero system of the nation-state. The safety coalition finds itself in the uncomfortable position of arguing for restraint in an environment where restraint is being redescribed as strategic surrender.
Stephen Turner‘s critique of essentialism explains why the fight between these coalitions never resolves. There is no single stable essence of authentic AI development being transmitted intact. There are competing reconstructions. One faction reconstructs the industry around safety and alignment density. Another reconstructs it around rapid capability growth and workable deployment. Both claim continuity with the original mission. Both select from the same dense world of research papers, scaling laws, and benchmarks to support present positions. What gets transmitted is not a stable essence but a body of material from which each coalition selects the passages and emphases that authorize its current stance.
Authority in this context is atmospheric. It lives in who gets platformed at major conferences, who secures the largest compute contracts, which labs are quietly recommended for top talent, and which ones are spoken of with hesitation. Minute variations in practice, whether a lab truly invests in alignment research or engages in safety theater, whether model releases follow internal safety commitments or are accelerated by competitive pressure, function as jurisdictional markers. They signal which authority structure a person has accepted as binding and which summons he or she is available to receive.
Across all three master domains, the same pattern holds. Safety traditionalists claim fidelity to uncompromising alignment standards. Accelerationists claim fidelity to sustainable progress under actual competitive conditions. Sovereign AI advocates claim fidelity to national interest framed as civilizational defense. None presents its position as interest-driven. All present it as what authentic AI stewardship requires. That convergence of form with divergence of content is precisely what Pinsof’s framework predicts. Moral language is the medium through which coalitions compete because it is the only language that converts a bid for institutional control into a legitimate claim on collective identity.
The jurisdictional war in the AI industry is a struggle over who gets to define what being summoned really requires. Beneath that, it is a struggle over which version of the hero system is strong enough to keep the terror contained. The expansion of AI into new labs, nations, and applications does not dissolve that internal tension. It amplifies it, because every new frontier lab or sovereign AI initiative that enters the serious coalition becomes a new arena in which the same question must be answered. How demanding must the safeguards be to remain credible? Where is the line between a field that sustains beneficial intelligence and an accommodation that hollows it out? The AI industry has been arguing over that line for years. The rest of technological civilization is now beginning to argue over it too, and unlike every other domain this series has examined, the outcome of that argument might determine whether there is a civilization left to do the arguing.
