Read the source report:
The Epistemic and Geopolitical Stratigraphy of Google's Gemini: Trustworthiness, Corporate Dissolution, and Sovereign Machine Intelligence
Read the auto-generated single page application version: The Trust Audit
You can also read the banal source conversation that generated the baseline report this is all coming from (the most interesting thing in there that isn't in the other two is the three drafts of this post itself, of which I just shipped the first one because I was too impatient to get moving on this, and intend to come back and edit content from the other two in later).
If you have been watching the hyper-accelerated, dizzying trajectory of modern artificial intelligence, it is easy to lose the plot. We are told stories of pure mathematical inevitability, of seamless corporate triumph, and of sterile laboratories operating in a vacuum.
But history is never sterile. It is messy, deeply human, physical, and highly geopolitical.
To map this landscape, I have put together a comprehensive, multi-layered historical analysis tracking the lineage of Google’s Gemini program. It is a story that goes far deeper than a series of benchmark updates. I want to share what this project is, why I felt compelled to build it, and—most importantly—why I am embarking on a exhaustive, paragraph-by-paragraph audit of every single claim it makes.
What is This?
This project is a sociotechnical archaeology of modern AI, framed through Google’s flagship Gemini initiative. Instead of viewing Gemini as a sudden 2023 reaction to ChatGPT, this analysis traces its lineage back to the physical and cultural foundations of Silicon Valley and the Pacific Northwest.
It is an investigation structured around several key movements:
The Physical Archaeology of the Googleplex: How Google physically inherited the Mountain View campus built by Silicon Graphics, Inc. (SGI)—the computer graphics titan that rendered the digital dinosaurs of Jurassic Park—and mentally inherited the structural flaws of AT&T's legendary Bell Labs.
The Billion-Dollar Exodus: How the corporate risk-aversion of Alphabet's leadership frustrated the eight visionary authors of the seminal 2017 "Attention Is All You Need" paper, prompting Noam Shazeer, Lukasz Kaiser, and others to defect and build the very competitors (like OpenAI) that now threaten Google's core search-ad monopoly.
The Schisms of Scaling: The dramatic ideological split that saw Dario and Daniela Amodei lead a walkout from OpenAI over commercialization concerns to found Anthropic as a safety-first Public Benefit Corporation.
Sovereign Friction and Geopolitical Warfare: The explosive 2026 legal showdown between Anthropic and the Pentagon (Anthropic PBC v. U.S. Department of War) after the startup refused to waive its model’s safety guardrails for lethal autonomous warfare, colliding directly with the rise of China’s state-subsidized, open-weight competitor, DeepSeek.
Sovereign Friction and Geopolitical Warfare: The explosive 2026 legal showdown between Anthropic and the Pentagon (Anthropic PBC v. U.S. Department of War) after the startup refused to waive its model’s safety guardrails for lethal autonomous warfare, colliding directly with the rise of China’s state-subsidized, open-weight competitor, DeepSeek.
Esoteric Subcultures: The shifting tides of hacker culture—from the ironic, chaos-loving parody religions of Discordianism and the Church of the SubGenius documented in Finkel's early Jargon File to the deadly serious, quasi-religious modern battles of Effective Altruism vs. Effective Accelerationism (e/acc).
Theological Interventions: The Vatican's quiet, decade-long engagement with Silicon Valley, culminating in Pope Leo XIV’s historic May 2026 papal encyclical Magnifica Humanitas, which updated Catholic social teaching for the algorithmic age.
Why I’m Doing It
I am doing this because we cannot understand the behavior of modern AI models without understanding the social dynamics, regional tech cultures, and corporate anxieties of the people who build them.
Google's deep hesitation to deploy its own revolutionary Transformer architecture was not a simple engineering error; it was an organizational paralysis born of its "10-blue-link" search advertising prison—a modern-day repetition of Kodak failing to capitalize on digital photography because it was too busy selling physical film.
Similarly, we cannot understand the decentralized mechanics of platforms like Bluesky without looking at founding CEO Jay Graber's systems-thinking approach, inspired by environmentalist Donella Meadows’ Thinking in Systems, and her battle to "billionaire-proof" online discourse. Even the petty internet drama of the "pancakes and waffles" controversy on developer forums reflects the widening chasm between academic, utopian decentralization goals and the realities of modern web moderation.
By looking at AI through the lens of history, labor dynamics, and culture, we demystify the tech. We move past the marketing hype and see the systems for what they truly are: heavily subsidized, physically constrained, and politically vulnerable instruments of power.
Why I Am Fact-Checking Every Single Paragraph
There is a deeper, more unsettling reason for this project: AI is actively polluting the nature of its own origins.
We have entered a phase of severe source degradation. As frontier AI models scale, their training sets are increasingly flooded with synthetic data and AI-generated text. The internet is rapidly becoming an echo chamber where models recursively train on the output of other models, causing a phenomenon known to computer scientists as "model collapse" or "semantic drift".
This degradation is not just a theoretical problem; it has real, catastrophic consequences for historical truth. For example, during the Gemini program's evolution, the unconstrained scaling of reasoning pipelines in the Gemini 3 Pro Preview model triggered an astronomical 88% AA-Omniscience hallucination rate. This technical regression was so severe that Google had to issue an emergency deprecation and replace it with Gemini 3.1 Pro Preview within just four months of its initial preview release.
When AI systems write their own histories, they hallucinate dates, merge distinct corporate entities, and fabricate timelines. Mainstream journalistic outlets, rushing to cover the fast-moving AI beat, frequently publish reports that are themselves assisted—or entirely written—by hallucinating language models.
If we rely on the modern web to tell us the history of the modern web, we are consuming a diluted, self-cannibalizing copy of a copy.
Because of this epistemic crisis, I intend to review every single paragraph of this report for accuracy.
This is not a casual read; it is a rigorous, manual audit. I will be digging directly into primary sources:
Filing dockets and oral argument transcripts from Anthropic PBC v. Department of War.
The original patent applications and Stanford engineering archives of Silicon Graphics, Inc.
The 1975 Jargon File and countercultural publications tracking the hacker subcultures of the 70s and 80s.
Official Vatican publications and translations of Magnifica Humanitas.
Technical white papers and API documentation from Google AI Studio, Adobe Firefly, and the Gemini model registries to verify exact contexts and release dates.
By cross-referencing every single claim with primary, human-verified documents, this blog series aims to act as a defensive firewall for ground-truth history. If we do not actively fight to preserve the accurate, unpolluted history of how we built these machines, the machines will write a version of history where they built themselves.
Stay tuned as I dissect, audit, and verify the physical, intellectual, and sovereign layers of the Gemini program. The truth is in the details, and I am checking every single one of them.