Money. Identity. Technology.
Follow the money. Find the decision.
AI supplier contesting policy, use conditions and government authority
Anthropic develops Claude and participates in AI policy, commercial supply and government deployment. Its place in this network combines a model provider’s market power, announced funding for public advocacy and a disputed relationship with the US defense establishment over the terms on which its technology may be used.
Anthropic’s role begins with Claude, a commercially supplied AI system. Its early financing also connects it to this investigation’s philanthropic network: the company says Jaan Tallinn led its $124 million round in 2021. In July 2026 it announced a total of $40 million in support for Public First Action, with restrictions on election spending. Model development and public-policy activity are both part of its institutional footprint.
Government deployment introduces other powerful actors. A November 2024 announcement described Claude integration through Palantir’s platform, Amazon SageMaker and AWS-supported infrastructure for US intelligence and defense operations. The pathway brings together a model developer, an integrator, a cloud provider and public customers with their own operational requirements.
The 2026 defense litigation clarifies where control sat in one disputed deployment. The court’s factual account describes Claude Gov’s use restrictions as contractual: Anthropic could not see the department’s prompts or technically intervene in the deployed model. The government controlled when future versions would be deployed. The ability to state a permitted-use condition and the ability to operate the system were held differently.
On August 27, the court vacated the challenged designation and broad contractor boycott and enjoined specified implementation, while preserving lawful government choices to move to other suppliers. That outcome makes this a useful example of contested power with an actual remedy. Anthropic can shape supply and advocate policy; government can purchase, demand and switch; courts can constrain how the public authority exercises those powers.
In September 2026, Dario Amodei committed Anthropic to ongoing outside reviewers with access comparable to internal risk staff. He proposed shared industry standards and government-supported coordination next. The planned reviewers could publish findings, with specified redactions; METR appears as an example, not an appointment. The contract, selected team and practical handling of access disputes remain the decisive missing records.
Assessment updated 2026-09-13
Anthropic → U.S. Department of War / Defense
Checked 12 September 2026
The court's factual account describes Claude Gov restrictions as contractual, without Anthropic visibility into DoW prompts or technical intervention in deployed behavior.
Anthropic → Anthropic embedded external-review commitment
Event or period: 2026-09; checked 13 September 2026
Amodei commits Anthropic to employee-like reviewer access and publication rights subject to specified redactions. No installed team or executed contract was established.
Greg Casar → Anthropic
2026-09-02
Casar's September 2 release says Anthropic answered August requests but withheld logs and sufficient answers; follow-up demanded fuller disclosure by September 15.
Anthropic v DoW: August 2026 final relief → Anthropic
Checked 12 September 2026
Docket 251 of 27 August 2026 vacates challenged measures and permanently enjoins specified implementation.
Anthropic → Palantir Technologies Inc.
Checked 12 September 2026
The November 2024 issuer release combines Claude, AIP, SageMaker and an AWS-supported IL6 environment.
Jaan Tallinn → Anthropic
Checked 12 September 2026
Anthropic says Tallinn led its $124m financing round announced 28 May 2021.
Rohan Paul → Anthropic
2026 publication; checked 12 September 2026
Circulates assessment. Does not establish employment or sponsorship.
Anthropic → Gates–Anthropic public-benefit AI partnership
Event or period: 2026-05-14; checked 14 September 2026
Credits and technical support within the same package.
A separate designation track under 41 USC 4713/FASCSA is identified in September 3 legal commentary. D.C. Circuit cases 26-1049/26-1162 remained in briefing activity in the public mirror last retrieved September 3, including the government's response to Anthropic's August 28 additional-authority letter. No merits disposition is visible in that snapshot. The public mirror is not a September 12 live-PACER record. D.C. Circuit docket 26-1049, consolidated26-1162 ; secondary explanation: Jones Walker analysis of parallel designations .
Read the research & sources ↗The accompanying four-page final-relief order, docket 251, vacates the section 3252 supply-chain designation and the broad prohibition on defense contractors doing business with Anthropic, and permanently enjoins implementation of the challenged actions against specified defendants. Paragraph 14 expressly preserves lawful actions, including choosing another AI provider; it does not require DoW to use Anthropic. Some claims/defendants were excluded, so this was not an unqualified victory on every claim. Anthropic v DoW order of final relief, docket 251 .
Read the research & sources ↗Anthropic’s account adds operational connections: interfaces, evaluation tools, health-ministry work and an IDM integration. These are worth following beyond the announcement.
Read the research & sources ↗The wording deserves scrutiny. It contains no named Anthropic or closed-model exemption. The proposed agency enforces; the board advises. Development and deployment are the stated targets; a blanket criminal offence for merely downloading an open model is not established.
Read the research & sources ↗Separate disclosed commercial placements from the people and stories Paul circulates.
Open diagram →Founder votes, purchasing power and AI distribution create different forms of authority around Karp's company.
Open diagram →Follow grants, lab access, people and public-sector work around the evaluator.
Open diagram →Follow the joint programme and its institutional connections.
Open diagram →Follow the named relationships between the main participants in this part of the investigation.
Open diagram →Anthropic → $40m total announced support → Public First Action
2026-07-21
Reported by the cited source
July 21, 2026 announcement restricts election spending. Not independently audited payment here; distinct from People First Marketing.
Greg Casar → Demands incident records; receives response → Anthropic
2026-09-02
Casar's September 2 release says Anthropic answered August requests but withheld logs and sufficient answers; follow-up demanded fuller disclosure by September 15. Company correspondence and criticism are established from legislator release; underlying response PDFs not read. No adviser/donor relationship implied.
Rohan Paul → Circulates assessment → Anthropic
2026 publication; checked 12 September 2026
Circulates assessment. Does not establish employment or sponsorship.
Jaan Tallinn → Led 2021 financing round → Anthropic
Checked 12 September 2026
Anthropic says Tallinn led its $124m financing round announced 28 May 2021. The $124m is the entire round, not Tallinn’s individual investment.
Anthropic → Claude integration announced → Palantir Technologies Inc.
Checked 12 September 2026
The November 2024 issuer release combines Claude, AIP, SageMaker and an AWS-supported IL6 environment. Historical integration announcement; current deployment continuity is not established after the 2026 dispute. No equity terms disclosed.
Anthropic v DoW: August 2026 final relief → Designation and boycott vacated → Anthropic
Checked 12 September 2026
Docket 251 of 27 August 2026 vacates challenged measures and permanently enjoins specified implementation. Not every claim or defendant; separate procurement-security review remains.
Anthropic → Contractual use limits; no technical veto → U.S. Department of War / Defense
Checked 12 September 2026
The court's factual account describes Claude Gov restrictions as contractual, without Anthropic visibility into DoW prompts or technical intervention in deployed behavior. Scoped to the court's described national-security deployment; not all Anthropic products. DoW controlled future deployment.
Anthropic → Commits to ongoing outside evaluator access → Anthropic embedded external-review commitment
Event or period: 2026-09; checked 13 September 2026
Amodei commits Anthropic to employee-like reviewer access and publication rights subject to specified redactions. No installed team or executed contract was established.
Joe Benton → Former Scalable Oversight manager → Anthropic
Checked 14 September 2026; event date unresolved
Also led Anthropic Fellows research.
Anthropic → Joint programme partner → Gates–Anthropic public-benefit AI partnership
Event or period: 2026-05-14; checked 14 September 2026
Credits and technical support within the same package.