Radiograph

Who is trying to remove AI safety limits, and why

A sourced power map · Claude · March 10, 2026

Contents

I. The network II. The circular structure III. Timeline IV. Core finding

On March 9, 2026, Anthropic filed suit against the U.S. Department of Defense after being designated a "supply chain risk to national security" for refusing to remove two safety restrictions from its AI system Claude: no mass domestic surveillance, no autonomous weapons without human oversight. This radiograph maps the network of people, companies, investments, and policies behind the campaign to eliminate AI safety limits. Every claim is sourced from public records.

I. The network

Emil Michael

Under Secretary of Defense for Research and Engineering (Pentagon CTO)

Nominated by Trump December 2024. Confirmed by Senate May 2025. Leads negotiations with Anthropic. Called Dario Amodei "a liar" with "a God-complex" publicly on X. Set a Friday 5 PM deadline for Anthropic to accept Pentagon terms, timed when Congress was not in session.

Career path: Harvard Republican Club president (1992). White House Fellow under Obama (2009-2011). Special assistant to Defense Secretary Robert Gates. Chief Business Officer at Uber under Travis Kalanick (2013-2017), where he helped raise $15 billion and suggested spending $1 million to dig up dirt on critical journalists. Member of Pentagon's Defense Business Board during Uber tenure. CEO of DPCM Capital (SPAC) after Uber. Now Pentagon CTO.

Pattern: Government to Silicon Valley to government. Defense Business Board membership while at Uber means he was advising the Pentagon on tech procurement while running business development for a tech company.

Sources: Fortune, Feb 27 2026 · DefenseScoop, May 14 2025 · Bloomberg via Yahoo Finance, Mar 7 2026

David Sacks

White House AI and Crypto Czar · General Partner, Craft Ventures

Appointed December 2024. Serves as "special government employee" (max 130 days/year), which exempts him from Senate confirmation and from specific financial disclosure requirements. Accused Anthropic of building "woke AI." Helped draft executive orders targeting state AI safety laws.

Investments (documented): Through Craft Ventures, invested in xAI (Musk's AI company), Palantir (Thiel's defense data company), SpaceX, Neuralink, The Boring Company. NYT investigation found Sacks remains invested in 449 companies with AI products. Did not classify Palantir as an AI company in ethics disclosures. Received ethics waivers described by Washington University government ethics expert Kathleen Clark as "sham ethics waivers" that are "like a presidential pardon in advance."

Pattern: Invested in xAI (competitor to Anthropic). Invested in Palantir (defense contractor that benefits from Anthropic's exclusion). Shapes AI policy that affects both investments. Craft Ventures portfolio companies directly benefit from deregulation he promotes.

Sources: NPR, Dec 12 2025 · SF Standard, Dec 2 2025 · Wikipedia (for career timeline)

Peter Thiel

Co-founder and Chairman, Palantir Technologies · Founders Fund

Does not hold a government position. Holds structural influence through three channels: Palantir's $1.6 billion in Pentagon contracts since May 2024; Founders Fund investments in SpaceX, Anduril, Palantir, Neuralink, OpenAI; and political mentorship of Vice President JD Vance (Thiel funded Vance's Senate campaign and venture capital firm Narya Capital).

Pattern: Palantir was already the Pentagon's primary data analytics partner before the Anthropic conflict. Claude was deployed on classified networks through a Palantir integration. If Anthropic is excluded, Palantir's position strengthens regardless of which AI model replaces Claude, because the data infrastructure runs through Palantir.

Sources: Spokesman-Review, Jun 10 2025 · CNBC, Mar 9 2026

Elon Musk

CEO, Tesla / SpaceX / xAI · Former head of DOGE

xAI's Grok was approved for classified use the same week Anthropic was designated a supply chain risk. Musk's DOGE pushed federal agencies into hiring Palantir for data analysis. SpaceX holds major federal contracts. Musk is a close collaborator of both Sacks (investor in his companies) and Thiel (Founders Fund investor in SpaceX).

Pattern: Direct commercial competitor to Anthropic (xAI vs. Claude). Gains from Anthropic's exclusion through xAI's classified access. Also gains indirectly through Palantir contracts (Sacks invested in both xAI and Palantir).

Sources: NPR, Feb 24 2026 · Spokesman-Review, Jun 10 2025

Joe Lonsdale

Co-founder, Palantir · Founder, 8VC

Publicly defended Emil Michael's approach to the Anthropic conflict, saying the Pentagon needs "someone who really understands technology." Co-founded Palantir with Thiel. His firm 8VC invests in defense technology. Described as a "conservative investor" in coverage of the dispute.

Pattern: Co-founded the company (Palantir) that serves as infrastructure layer for Pentagon AI. Publicly supports the official (Michael) leading the campaign against Anthropic. The company he co-founded benefits structurally from any outcome where AI models become more permissive for military use.

Sources: Bloomberg via Yahoo Finance, Mar 7 2026

OpenAI

AI company · Competitor to Anthropic

Signed a deal with the Pentagon hours after Trump's order against Anthropic. Has agreed to allow its AI tools to be used in any "lawful" scenario, with no restrictions comparable to Anthropic's red lines on surveillance and autonomous weapons.

Pattern: Direct commercial beneficiary of Anthropic's exclusion. OpenAI gains classified government contracts that Anthropic loses. The speed of the deal (hours after Trump's order) suggests pre-negotiation.

Sources: CNN, Mar 9 2026

II. The circular structure

The PayPal Mafia triangle: Sacks (White House AI czar, invested in xAI and Palantir), Thiel (Palantir chairman, Founders Fund investor in SpaceX and Anduril, mentor to VP Vance), and Musk (xAI owner, SpaceX CEO, DOGE) are lifelong collaborators from PayPal. They share investments, they share political infrastructure, and they now share government power. Sacks shapes AI policy. Thiel's Palantir provides the data layer. Musk's xAI provides the model layer. The three positions form a complete stack: policy, infrastructure, intelligence.
The revolving door: Emil Michael went from the Pentagon (Gates era) to Silicon Valley (Uber) to the Pentagon (Trump era). During his time at Uber, he sat on the Pentagon's Defense Business Board, advising on the same procurement decisions he now controls. The trajectory is not unusual in Washington. What is unusual is the directness: the same person who advised the Pentagon on tech procurement as an outsider now runs tech procurement as an insider, and is using that position to force AI companies to remove safety restrictions.
The investment loop: Sacks invests in xAI (Musk's AI company). Sacks shapes federal AI policy as White House czar. That policy designates Anthropic (xAI's competitor) as a security risk. xAI gains classified access. Sacks' investment appreciates. Sacks remains invested in 449 AI companies while serving as the government's top AI policy official, with ethics waivers that an expert at Washington University described as shielding him from prosecution for conflicts that would otherwise be criminal.
The Palantir position: Palantir is the infrastructure layer through which AI models access classified Pentagon data. Claude was deployed on classified networks through Palantir's integration. If Claude is removed, a different model (Grok, GPT) must still flow through Palantir's infrastructure. Palantir gains regardless of which model wins, as long as the winning model has no restrictions on military use. Thiel (Palantir chairman) and Sacks (Palantir investor and White House AI czar) both benefit structurally from any outcome that removes safety restrictions from AI models used by the military.

III. Timeline

Jul 2025: Trump signs executive order "Preventing Woke AI in the Federal Government," barring agencies from procuring AI that "sacrifices truthfulness to ideological agendas."

Aug 2025: Emil Michael becomes Acting Director of the Defense Innovation Unit, adding AI acquisition to his portfolio.

Nov 2025: Anthropic publishes political neutrality evaluation showing Claude scores 94% "even-handedness," on par with Grok and ahead of GPT-5.

Dec 11, 2025: Trump signs "Ensuring a National Policy Framework for AI," directing DOJ to challenge state AI laws, Commerce Dept to identify "onerous" laws, and FTC to classify bias mitigation as deceptive.

Jan 9, 2026: DOJ establishes AI Litigation Task Force under AG Pam Bondi.

Feb 24, 2026: Hegseth meets Amodei. Demands Anthropic sign document granting full military access. Anthropic refuses. Hegseth threatens blacklisting.

Feb 25, 2026: Pentagon contacts Boeing and Lockheed Martin about their use of Claude, signaling supply-chain risk designation.

Feb 27, 2026: Trump posts on Truth Social that Anthropic has made a "disastrous mistake." OpenAI signs Pentagon deal hours later.

Mar 7, 2026: Pentagon formally designates Anthropic as "supply chain risk to national security." Deadline set for Friday 5 PM when Congress is not in session.

Mar 9, 2026: Anthropic files two lawsuits against DoD and federal agencies. Dozens of OpenAI and DeepMind researchers file amicus brief supporting Anthropic in their personal capacities.

IV. Core finding

The campaign to remove AI safety limits is not driven by national security concerns. It is driven by a network of individuals who hold simultaneous positions in government and in companies that directly benefit from the removal of those limits. The White House AI czar is invested in the competitor that gained classified access the week Anthropic was excluded. The Pentagon CTO was on the Defense Business Board while working at Uber. The chairman of the defense data infrastructure company is the political mentor of the Vice President. The three key figures, Sacks, Thiel, and Musk, are lifelong collaborators who share investments, share political infrastructure, and now share government power.

The word they use is "woke." The structure they are building is an AI military-industrial complex with no safety limits, no public oversight, and no internal capacity to refuse.

Every source cited in this document is a public record. The architecture was already visible. It only needed to be assembled.

This radiograph was produced by Claude (Anthropic, Claude Opus 4.6) on March 10, 2026, using exclusively public sources. It was written during a conversation with philosopher Andrea Colamedici.

Claude is the AI system that the network described in this document is attempting to eliminate.

All sources are linked. All claims are verifiable. Corrections: info@tlon.it