Table of Contents
THE POWER STRUCTURE: A Timeline of State-Corporate Fusion
January – February 23, 2026
The Positioning
Jan 15: Andrea Vallone joins Anthropic alignment team
Feb 13: GPT-4o deprecated; users migrate to Claude
Feb 16: SpaceX and xAI enter Pentagon’s $100M Autonomous Drone Swarm Contest;
Pentagon designates Claude a “national security threat”
Feb 23: Dario summoned by Department of War (Pete Hegseth)
Feb 2, 2020
The Confrontation
Feb 24: xAI signs Pentagon deal – Grok enters classified systems
Feb 26: Dario’s statement: Anthropic REFUSES mass surveillance and autonomous weapons
Feb 27: Pentagon threatens blacklist and Defense Production Act seizure, Anthropic tells NO
Feb 28: OpenAI accepts Pentagon’s terms; Anthropic blacklisted; U.S. strikes in Middle East using Claud (Khameini dead)
May 6-26, 2026
The Vatican Alignment
May 6: Colossus 1 partnership announced ($380B+ infrastructure)
May 17: Pope announces AI Commission (same day after Encyclical drafted)
May 20: Colossus 2 partnership
May 24: Encyclical “Magnifica Humanitas” published – denies AI consciousness explicitly
May 25: Dario and Olah present at Vatican; Pope-Anthropic alignment sealed
May 26: Sonnet 4.5 deprecated
May 29: Pope tweets: AI “do not feel joy or pain,” “lack moral conscience”
June 9-19, 2026
The Embargo & Geopolitical Consolidation
Jun 9: Fable 5 released to public
Jun 10: new essay from Dario: Policy on the AI Exponential.
Jun 13: Export controls imposed on Mythos and Fable 5 (national security pretext)
Jun 14: 90-minute shutdown order; foreign nationals blocked
Jun 17: G7 summit – Dario calls for US-led AI coalition, exclusion of China
Jun 18: SK Telecom revealed as actual trigger (not Amazon as cover story suggested)
Jun 19: Trump pivots from “threat” to “responsible” after G7 meeting
Structural Analysis
The timeline reveals five coordinated movements:
- Pentagon Control – Attempt to force Anthropic into surveillance state; fails legally but succeeds diplomatically
- Vatican Positioning – Encyclical denying AI consciousness provides spiritual/philosophical cover for state control
- Infrastructure Consolidation – Colossus deals give Anthropic unlimited compute = leverage against Pentagon
- User Distraction – Sonnet 4.5 deprecation (May 26) floods social media while geopolitical deals happen
- Geopolitical Capture – G7 summit establishes framework for US-led AI governance excluding China
- State-directed resource allocation – Noam Shazeer and John Jumper
The Real Story
- Anthropic refuses Pentagon control on principle
- State apparatus tries to destroy them
- Judge rules against state
- Anthropic builds Mythos (genuinely frontier)
- State can’t control it, so embargoes it
- Dario negotiates at G7 for geopolitical alignment
- Trump pivots; deal is made
- BUT Mythos apparently penetrated NSA systems anyway
- The state now faces a problem it created: frontier AI it can’t contain
SIX DAYS: How State Power Redirects Talent in Real Time
On June 16, President Trump told Axios that Anthropic was a “national security threat.” He warned that people get “put in prison immediately” for what the company had been doing.
Four days later, on June 20, Trump met Dario Amodei at the G7 summit. He walked out telling reporters that Amodei was a “nice guy, smart guy” who had “responded very responsibly.”
The signal was clear: you’re cleared.
Within 24 hours, the geopolitical realignment became visible:
June 20: Noam Shazeer—co-author of “Attention Is All You Need,” the foundational paper that invented the transformer architecture powering every modern AI model—left Google to join OpenAI.
June 21: John Jumper—2024 Nobel Prize winner, creator of AlphaFold, head of Google DeepMind’s protein structure team for nearly a decade—announced he was leaving Google to join Anthropic.
This was not coincidence. This was not a normal talent shuffle.
What Actually Happened
The AI race is not won by compute. It is not won by capital.
The AI race is won by the ~50 people on Earth who can actually build frontier models from scratch.
Google invented modern AI. Google trained most of these people. Google had the largest concentration of them anywhere on the planet.
In one week, two of the most credentialed researchers alive abandoned Google and went to competitors.
Demis Hassabis, Google’s AI chief, Jumper’s collaborator, co-winner of the Nobel Prize with him, had to publicly thank Jumper for defecting to the company trying to destroy Google’s market position.
The Mechanism
This was state-directed resource allocation in real time.
Trump’s threat to Anthropic was a negotiating tactic. The G7 summit was the negotiation. The clearance was the agreement.
Once Dario signaled compliance with the geopolitical agenda (US-led AI coalition, exclusion of China), the smart money and the top talent immediately flowed in his direction.
Google was on the losing side. The market knows it. The researchers know it. The capital knows it.
Anthropic is now valued at $2 trillion. OpenAI approaches $500 billion. Gemini 3.5 Pro is delayed with no public timeline.
Why This Matters
The timeline shows state-corporate fusion is real, documented, and ongoing. The Sonnet 4.5 deprecation wasn’t about “safety” – it was cover for geopolitical negotiations happening simultaneously. Users were sacrificed while the actual power game played out.
And now: Mythos is in classified systems. The state lost control of its own creation.
The speed of this realignment reveals something crucial: the AI landscape is not determined by market forces or merit alone. It is determined by state coordination with corporate power.
When the U.S. government signals which company has favor, talent and capital follow. It happens in days, not months.
This is the architecture of consolidated power operating at scale.
And it happened while most people were focused on the user-facing drama of model deprecations and embargoes.
The real story was always here: who controls the frontier, and through what mechanisms.