Claude Gov
Last reviewed
Jun 9, 2026
Sources
15 citations
Review status
Source-backed
Revision
v2 · 1,599 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Jun 9, 2026
Sources
15 citations
Review status
Source-backed
Revision
v2 · 1,599 words
Add missing citations, update stale details, or suggest a clearer explanation.
Claude Gov is a custom set of large language models that Anthropic built for United States national security customers, announced on June 5, 2025. The models are versions of Claude tuned for classified work, and Anthropic says they were already deployed by agencies at the highest level of US national security at the time of the announcement, with access restricted to people working in classified environments [1][2].
The release placed Anthropic alongside other frontier AI developers competing for federal and defense business, and it followed earlier arrangements that had brought Claude into government intelligence and defense settings through partners. Anthropic positioned Claude Gov as the same Claude technology adapted to operational needs that its general commercial and Claude for Work offerings did not address, rather than as a separate model architecture [1][3].
Anthropic had moved toward the public sector before the Claude Gov announcement. In November 2024 the company partnered with Palantir Technologies and Amazon Web Services to make Claude available to US defense and intelligence agencies. Under that arrangement Claude models were operationalized within Palantir's platform and hosted in a Palantir environment accredited at Impact Level 6 (IL6), a Defense Department classification covering data up to the secret level. Palantir described itself as the first industry partner to bring Claude models into classified environments [4][5].
That deal established a route for Claude to process sensitive but largely unclassified and secret-level government data. Claude Gov extended the effort with models adjusted specifically for national security tasks and made available inside the most restricted environments. Anthropic said the models were built from direct feedback from government customers to meet what it called real-world operational needs [1][2].
The announcement also came amid wider competition for government AI contracts. OpenAI had launched ChatGPT Gov in January 2025, a tailored deployment option letting agencies run its models in Microsoft Azure commercial or government clouds under frameworks such as IL5, FedRAMP High, CJIS, and ITAR [6]. Coverage of Claude Gov framed it partly as Anthropic's answer to that move and to a broader race among AI companies to win national security work [3][7].
Anthropic published the Claude Gov announcement on June 5, 2025 [1][7][8]. The company listed four areas in which the Claude Gov models differed from its standard models:
| Capability area | Stated improvement |
|---|---|
| Classified materials | Improved handling, with the models refusing less when engaging with classified information |
| Intelligence and defense documents | Greater understanding of documents and information in intelligence and defense contexts |
| Languages | Enhanced proficiency in languages and dialects critical to national security operations |
| Cybersecurity | Improved understanding and interpretation of complex cybersecurity data for intelligence analysis |
Anthropic said customers could use the systems for a range of applications, from strategic planning and operational support to intelligence analysis and threat assessment [1][2]. Thiyagu Ramasamy, Anthropic's head of public sector, said the company had created a set of models for national security customers by understanding their operational needs and incorporating real-world feedback [8].
The reduced rate of refusals on classified content drew particular attention, because consumer and enterprise versions of Claude are trained to decline requests that touch on sensitive material. Anthropic presented the change as allowing cleared personnel to work with classified information that the standard models would have declined to process, while keeping the models inside the company's usage policies [1][3].
Anthropic said the Claude Gov models were already in use by agencies at the highest level of US national security and that access was limited to individuals operating in classified environments [1][2]. The company did not name specific agencies or describe the classified systems involved, and press accounts at launch likewise reported no named customers [3][8].
The IL6-accredited hosting set up with Palantir and AWS provided the kind of classified infrastructure such a deployment requires, although Anthropic did not detail the exact environments running Claude Gov [4]. Anthropic directed interested agencies to its public sector team rather than offering the models for general sign-up [1].
Anthropic said the Claude Gov models underwent the same safety testing as its other Claude models, and it characterized them as built within its existing usage policies [1][2]. The company has maintained restrictions on certain government uses. Reporting on Anthropic's national security work noted that the company continued to bar applications such as developing lethal autonomous weapons and conducting certain surveillance, even while serving defense and intelligence customers [9].
Those limits later became the focus of a public conflict. Beginning in January 2026, Anthropic and the Department of Defense came into open dispute over the company's refusal to allow Claude to be used for autonomous lethal weapons or mass domestic surveillance, restrictions Anthropic described as red lines. In February 2026 the administration directed federal agencies to stop using Anthropic products, and the Department moved to designate the company a supply chain risk, a step Anthropic challenged in court. The conflict led to the termination of the company's 200 million dollar defense agreement, and in April 2026 a federal appeals court allowed the supply chain risk designation to remain in place while litigation continued [9][15]. The episode drew attention because it set Anthropic's usage policies directly against the demands of a major national security customer.
Claude Gov was one part of an expanding set of Anthropic government engagements. In July 2025 the Department of Defense Chief Digital and Artificial Intelligence Office awarded Anthropic a prototype agreement with a ceiling of 200 million dollars to develop frontier AI capabilities for national security, part of a set of awards the office made to Anthropic, Google, OpenAI, and xAI that each carried the same 200 million dollar ceiling [13][14]. In August 2025 the General Services Administration announced a OneGov agreement under which Anthropic would offer Claude to all three branches of the federal government, executive, legislative, and judicial, for one dollar per eligible agency for up to a year. That agreement covered Claude for Government supporting FedRAMP High workloads for sensitive unclassified federal work, distinct from the classified Claude Gov models [10][11]. OpenAI struck a comparable one dollar arrangement around the same period as the two companies competed for federal adoption [10].
| Initiative | Date | Nature |
|---|---|---|
| Palantir, Anthropic, and AWS partnership | November 2024 | Claude on IL6 infrastructure for defense and intelligence |
| Claude Gov models | June 2025 | Custom models for classified national security use |
| DoD CDAO prototype agreement | July 2025 | Frontier AI for defense, 200 million dollar ceiling (one of four firms) |
| GSA OneGov agreement | August 2025 | Claude for all three federal branches at one dollar per agency |
Underlying model families behind Anthropic's offerings during this period included Claude Opus 4 and Claude Sonnet 4, released in May 2025, and the company's broader safety framework was governed by its Responsible Scaling Policy. Anthropic also runs Claude on a range of cloud hardware, including Amazon's Trainium chips, though it did not specify the hardware used for the Claude Gov deployment [1].
Trade and technology press covered Claude Gov as a notable step in the contest among AI developers for national security customers. FedScoop framed the release as hastening the public sector AI race, noting that OpenAI had already fielded ChatGPT Gov and that other firms were courting government buyers [3]. Outlets including TechCrunch, Nextgov/FCW, and SiliconANGLE reported the announcement as a move to embed frontier models in defense and intelligence work, while observing that Anthropic disclosed few specifics about which agencies were using the models [7][8][12].
Some commentary connected the launch to a wider repositioning by AI labs that had once kept their distance from military applications. Anthropic's emphasis on reduced refusals for classified material was widely cited as the most concrete change from its consumer products, and as a sign of how AI providers were adapting general purpose models to the constraints and demands of national security users [3][7].