# Anthropic The AI safety company that became a political flashpoint. Founded in 2021 by former [[OpenAI]] researchers [[Dario Amodei]] and [[Daniela Amodei]], Anthropic built its brand on "responsible AI development" — then discovered that in Trump's Washington, responsibility is a liability. ## Founding & Origin Anthropic was born from OpenAI's internal tensions over safety vs. commercialization. Dario Amodei, OpenAI's VP of Research, left in 2021 along with ~10 senior researchers who believed OpenAI was moving too fast and prioritizing deployment over safety. The new company's stated mission: build AI systems that are "safe, beneficial, and understandable." **Key early investors with KB connections:** - **[[Jaan Tallinn]]** — Skype co-founder, AI existential risk mega-funder, TESCREAL institutional architect. Series A investor and board observer. - **Google** — $300M initial investment (2022), $2B more in 2023. Created an awkward competitor-investor dynamic. - **Amazon** — Up to $4B committed. Claude models hosted on AWS. The pipeline that led to the [[Palantir]]-Anthropic-Amazon Defense AI Deal]]. - **[[Dustin Moskovitz]]** — Facebook co-founder, [[Open Philanthropy]] / EA funder. Part of the [[TESCREAL]] funding network. ## The Pentagon Standoff (2025–2026) The defining event for understanding the political dynamics of the AI industry: ### Phase 1: The Contract (July 2025) Anthropic signed a **$200M Pentagon contract** with explicit restrictions: Claude would **not** be used for mass domestic surveillance of American citizens and would **not** power fully autonomous weapons. Claude Gov models were deployed at the highest levels of U.S. national security. Anthropic was the Pentagon's **preferred AI vendor** for classified work. ### Phase 2: The Demand (January 2026) Defense Secretary [[Pete Hegseth]] issued an AI strategy memo requiring **"any lawful use"** language in all DoD AI contracts — meaning no company-imposed restrictions. Anthropic refused to remove its guardrails. ### Phase 3: The Ultimatum (February 24, 2026) Hegseth gave [[Dario Amodei]] a deadline to comply. Anthropic held firm on two red lines: no mass surveillance of Americans, no fully autonomous weapons. ### Phase 4: The Ban (February 27, 2026) Trump ordered a **government-wide ban** on Anthropic products. Hegseth designated Anthropic a **"supply chain risk."** Trump publicly called the company **"Leftwing nut jobs."** The same evening, [[OpenAI]] announced its own classified Pentagon contract — with no usage restrictions. ### Phase 5: The Fallout - [[xAI]], [[Google DeepMind]], and OpenAI all accepted the "all lawful use" standard - Anthropic lost access to all federal contracts and classified systems - [[Center for American Progress]] called for Congress to establish AI procurement standards rather than leaving them to executive whim - EFF argued the episode showed that "privacy protections shouldn't depend on the decisions of a few powerful people" - Atlantic Council called it a "crisis of trust over AI" ### Phase 6: The Injunction and the Replacement Deals (March–May 2026) **March 26, 2026:** Federal Judge Lin (California) granted Anthropic a **preliminary injunction** blocking the government from enforcing the ban and the "supply chain risk" designation. The legal basis: the government likely violated the Administrative Procedure Act and [[First Amendment]] protections by retaliating against Anthropic for its speech (refusing to accept "all lawful use" terms). **April 17, 2026:** White House Chief of Staff [[Susie Wiles]] and senior officials met with Dario Amodei at the White House. Trump told CNBC a deal was "possible." The meeting signaled that Anthropic's legal position — and market position (Claude remained the #1 AI app) — forced a softening. **May 1, 2026:** The Pentagon announced **classified AI deals with seven companies** — SpaceX, [[OpenAI]], [[Google]], Nvidia, Reflection AI, Microsoft, and Amazon Web Services — to deploy AI on **Impact Level 6 and Impact Level 7 classified networks** via **GenAI.mil**, the military's central AI platform (already used by 1.3M+ personnel). **Anthropic was explicitly excluded.** The contracts use the "lawful operational use" language that Anthropic refused — deliberately replacing Anthropic's safety restrictions with a blanket authorization. The May 1 announcement completed the structural realignment: every major AI company except Anthropic has now accepted unrestricted military use as the price of federal contracts. The Pentagon's framing — "preventing AI vendor lock" and "ensuring long-term flexibility" — is the institutional language for the strategic message: any company that sets safety limits on military AI use will be replaced by one that does not. The "supply chain risk" designation remains in effect despite the injunction, creating ongoing legal ambiguity for defense contractors considering work with Anthropic. The episode established a precedent that extends beyond AI: federal contracts are now explicitly conditioned on the vendor's willingness to accept unrestricted government use. The chilling effect on corporate safety commitments across the tech industry is the point, not the side effect. ## Political Positioning Anthropic occupies an unusual position in the AI political landscape: it is neither fully aligned with the tech-libertarian right (it supports some AI regulation) nor with the progressive left (it still builds frontier AI and took defense contracts). **David Sacks conflict:** Trump AI Czar [[David Sacks]] publicly accused Anthropic of "running a sophisticated regulatory capture strategy based on fear-mongering" (October 2025). Amodei responded that managing AI risks should be "policy over politics." **State-level regulation:** Anthropic supported California's **SB 53**, a light-touch safety bill requiring frontier AI developers to publish safety protocols — putting it at odds with most of Silicon Valley and the Trump administration. **Dario Amodei's framing:** Called on Congress to legislate AI guardrails rather than leaving life-and-death decisions about surveillance and autonomous weapons to private companies or executive orders. The irony: by refusing to cede control, Anthropic was simultaneously arguing that companies shouldn't have this power. ## The Impossible Position The Anthropic saga illustrates a structural problem in the AI industry's relationship with government: 1. **Refuse military work entirely** → Lose revenue, lose influence over how AI is used in warfare 2. **Accept military work with restrictions** → Government retaliates when restrictions become inconvenient (this is what happened) 3. **Accept military work without restrictions** → Become complicit in whatever uses the government chooses (this is what OpenAI and xAI chose) There is no "safe" option. The companies that maintained ethical red lines were punished. The companies that didn't were rewarded with contracts. ## TESCREAL Connections Anthropic sits awkwardly within the [[TESCREAL]] framework. Its founders came from the AI safety community (rationalist-adjacent), its early funders include TESCREAL mega-donors ([[Jaan Tallinn]], Dustin Moskovitz), and its core concern — existential risk from AI — is a longtermist priority. But its willingness to accept regulation and refuse unrestricted military use puts it at odds with the accelerationist wing of the bundle. ## Development Log | Date | Development | Source | |------|------------|--------| | **Jul 2025** | $200M Pentagon contract signed with safety restrictions (no mass surveillance, no autonomous weapons) | CNBC, CNN | | **Jan 2026** | Hegseth issues "any lawful use" memo; Anthropic refuses | PBS, CNN | | **Feb 24, 2026** | Hegseth ultimatum to Amodei; Anthropic holds red lines | CNN | | **Feb 27, 2026** | Trump orders government-wide ban; Anthropic designated "supply chain risk" | CNBC, multiple outlets | | **Mar 26, 2026** | Federal Judge Lin grants preliminary injunction blocking ban | Court filing | | **Apr 17, 2026** | White House meeting: Chief of Staff Susie Wiles and officials meet Dario Amodei; Trump says deal "possible" | CNBC | | **May 1, 2026** | Pentagon announces classified AI deals with **7 companies excluding Anthropic** (SpaceX, OpenAI, Google, Nvidia, Reflection AI, Microsoft, AWS); GenAI.mil platform for IL-6/IL-7 networks. Pentagon framing: "never again rely on single AI provider" | Breaking Defense, TechCrunch | | **Early May 2026** | Signs of rapprochement: **Mythos Preview** (Anthropic's latest model) reportedly drawing Pentagon interest despite ban. Former **[[RAND Corporation]] head joins Anthropic**, signaling defense establishment credibility-building. Injunction remains in effect, blocking "supply chain risk" designation enforcement | Multiple outlets | | **Jun 19, 2026** | **Nobel laureate [[John Jumper]] leaves [[Google DeepMind]] for Anthropic.** The [[AlphaFold]] co-creator and 2024 Chemistry laureate is the highest-profile research hire of Anthropic's talent run; aligns with Anthropic's push into life sciences / computational biology. Paired with [[Noam Shazeer]]'s exit to [[OpenAI]] days earlier, the dual loss helped erase $200B+ from [[Alphabet]]'s market cap | CNBC, TechCrunch | ## Next Steps to Track - **Injunction appeal** — government may appeal Judge Lin's preliminary injunction - **Deal negotiations** — Trump-Amodei White House meeting signals possible resolution; terms unclear - **Mythos Preview** — Pentagon interest in latest model despite ban; potential workaround via injunction - **IL-6/IL-7 deployment** — 7 competitors deploying on classified networks without Anthropic - **RAND hire implications** — defense establishment figure joining Anthropic may signal strategic pivot - **Supply chain risk designation** — blocked by injunction but not withdrawn; legal limbo continues ## Talent Acquisition Even while banned from federal contracts and litigating against the Trump administration, Anthropic became a primary destination for elite AI research talent across 2025–2026. The signature catch came June 19, 2026, when Nobel laureate [[John Jumper]] — co-creator of [[AlphaFold]] — left [[Google DeepMind]] to join the company, signaling intent to compete in AI-for-science beyond large language models. Several of his AlphaFold collaborators were reported to be following him. The hire landed the same week [[Noam Shazeer]] left DeepMind for [[OpenAI]], and the twin departures wiped $200B+ off [[Alphabet]]'s market value. ## See Also - [[Dario Amodei]] - [[John Jumper]] — Nobel laureate hire (June 2026) - [[Palantir-Anthropic-Amazon Defense AI Deal]] - [[OpenAI]] — Primary competitor; accepted Pentagon's "all lawful use" standard - [[xAI]] — Also accepted "all lawful use" standard - [[Google DeepMind]] — Competitor on Pentagon AI contracts - [[Jaan Tallinn]] — Early investor, TESCREAL institutional architect - [[David Sacks]] — Trump AI Czar who accused Anthropic of "fear-mongering" - [[Pete Hegseth]] — Defense Secretary who issued the ultimatum - [[Dual-Use AI]] - [[AI Political Spending Ecosystem]] - [[TESCREAL]]