# Anthropic Contracts Government contracts, lobbying activity, and procurement dynamics for [[Anthropic]]. The company's contract history is the inverse of [[OpenAI Contracts]]: Anthropic had the Pentagon's preferred AI contract, refused to remove safety guardrails, lost everything, and is now fighting to get back in while simultaneously preparing for a trillion-dollar IPO. ## Contract table | Contract | Value | Date | Status | Notes | |----------|-------|------|--------|-------| | Pentagon classified AI | $200M | Jul 2025 | **Terminated** (Feb 2026) | Claude Gov deployed at highest classification levels. Red lines: no mass surveillance of Americans, no fully autonomous weapons. Contract killed when Anthropic refused "any lawful use" language | | [[Palantir-Anthropic-Amazon Defense AI Deal]] | Undisclosed | 2024 | **Frozen** | Three-way deal: [[Palantir]] integration layer, Anthropic models, Amazon AWS hosting. Frozen after government ban | | GenAI.mil platform | — | May 2026 | **Excluded** | Pentagon awarded classified AI deals to 7 companies (SpaceX, [[OpenAI]], [[Google]], Nvidia, Reflection AI, Microsoft, AWS) for IL-6/IL-7 networks. Anthropic explicitly excluded | ## Stated red lines Anthropic's two non-negotiable conditions for military contracts: 1. **No mass surveillance of American citizens** — Claude would not be used for domestic mass surveillance programs 2. **No fully autonomous weapons** — Claude would not power weapons systems that select and engage targets without human oversight These red lines are what triggered the government ban when [[Pete Hegseth]] demanded "any lawful use" language in all DoD AI contracts. **The analytical question:** were these red lines genuine safety commitments or strategic positioning? Anthropic accepted a $200M Pentagon contract for classified work — the red lines constrained *what kind* of military use, not *whether* there would be military use. The company's position was "responsible defense AI," not "no defense AI." ## Legal status The FASCSA supply chain risk designation creates a binary for defense contractors: - Contractors **must certify** they don't use Anthropic's Claude in military work - The designation remains in effect despite Judge Lin's broader injunction - D.C. Circuit denied Anthropic's emergency stay (Apr 8, 2026) but ordered expedited appeal - **Net effect:** Anthropic is locked out of DoD contracts but can still work with other government agencies ## Government staffing pipeline | Person | Role | Significance | |--------|------|-------------| | Jack Clark | Co-Founder, Head of Policy | Former [[OpenAI]] Policy Director; leads Anthropic's government affairs and regulatory strategy | | Chris Liddell | Board member | Former Deputy White House Chief of Staff (Trump first term); appointed to Anthropic board Feb 2026, the month of the ban | | Former RAND Corporation head | Unnamed hire | Defense establishment credibility-building; reported early 2026 | ## Lobbying spend | Year | Federal spending | Notes | |------|-----------------|-------| | 2023 | ~$300K | Initial Washington presence | | 2024 | ~$780K | Ramping; AI regulation focus | | 2025 | $3.13M | 4x increase; AI regulation, national security, export controls, procurement, infrastructure | | 2026 Q1 | $1.6M | 344% YoY increase for the quarter; **outspent [[OpenAI]] for the first time** ($1.6M vs. $1.5M) | **Focus areas:** AI safety regulation, federal procurement standards, export controls, national security AI policy, infrastructure. **Registered lobbyists** include Anthropic PBC's in-house team plus external firms. Lobbying topics span both defensive (fighting the ban) and offensive (pushing AI guardrails legislation like California SB 53). ## Political spending | Recipient | Amount | Date | Notes | |-----------|--------|------|-------| | Trump Inaugural Fund | $50K | Jan 2025 | 1/20th of what [[OpenAI]], Perplexity, and C3.ai each gave ($1M each) | | Public First Action | $20M | Feb 2026 | Largest AI-company political donation targeting safety regulation. Backing 30–50 candidates (both parties) in state and federal races through midterms | The $50K inaugural donation was a political hedge — enough to claim participation, small enough to signal that Anthropic wasn't buying access. The $20M to Public First Action was a direct challenge: using corporate money to elect candidates who would legislate the safety guardrails the administration was trying to eliminate through executive action. ## Procurement pattern analysis Anthropic's procurement trajectory is the mirror image of [[OpenAI Contracts]]: 1. **Build credibility through safety** → Position as the "responsible" AI vendor 2. **Win classified contract with guardrails** → Pentagon preferred vendor status (Jul 2025) 3. **Refuse to remove guardrails** → Government retaliates (Feb 2026) 4. **Litigate and lobby** → Fight the ban through courts and Congress simultaneously 5. **IPO as leverage** → A near-trillion-dollar public company is harder to blacklist than a private startup **The strategic bet:** Anthropic is wagering that the combination of legal victories, political spending, commercial market dominance (Claude #1 app), and IPO-scale market cap will eventually make the ban unsustainable. The alternative scenario: the defense market permanently bifurcates between "unrestricted" vendors ([[OpenAI]], [[xAI]], [[Google]]) and "excluded" vendors (Anthropic), and the safety brand becomes a commercial asset but a government liability. ## See Also - [[Anthropic Dossier]] — the document collection front door - [[Anthropic Watch Log]] — running OSINT feed - [[OpenAI Contracts]] — the competitor's contract ledger (accepted "all lawful use") - [[Palantir Contracts]] — the defense-tech archetype - [[Sole Source]] — procurement concept - [[Vendor Lock-In]] — the dynamic the Pentagon cited when replacing Anthropic - [[Lobbying]] — the concept