Anthropic Brings Claude Code and Claude Cowork to US Government

Anthropic launched a public beta of its Claude Code and Claude Cowork tools for the US federal government on July 7. Both products now run inside the Claude for Government Desktop app, in an environment authorized under FedRAMP High, the highest security certification level for cloud services used by US federal agencies.

Claude Code is meant to let public-sector teams build and modernize the IT systems behind government services. Claude Cowork, in turn, works directly on files stored on an official's computer, taking over tasks such as drafting internal memos, reviewing RFP submissions, or preparing presentations.

What Agencies Are Getting

Both tools are built on the same application used by Anthropic's commercial customers, but are delivered through an environment that meets the strict security requirements of the federal government. Conversation history stays local on devices managed by the agency, and deployment runs through the standard MDM platforms used across the public sector.

Department administrators can set default configurations, allocate budgets to individual subordinate units, and impose model and spending limits based on a given team's access level. SCIM group mapping allows access segmentation, so that different offices within the same agency can have different permission levels.

Security and Accountability

A key part of the offering is hash-chained audit logging, which prevents undetected tampering with records of system activity. Operations deemed sensitive require approval from two people, and usage data exports contain only billing information, never the content of conversations themselves.

Anthropic has also published a publicly available secure configuration guide compliant with FedRAMP requirements and provides formal change notifications under a non-disclosure agreement. A summary of penetration test results is available through the company's trust center, intended to speed up the Authority to Operate (ATO) process for individual agencies.

Billing and Business Model

Agencies can choose between standard license seats and custom packages with spending and model limits. Usage is purchased in pre-set packages with a hard cap that cannot be exceeded, intended to make budget planning easier for public bodies, where every expense is subject to review.

Anthropic acts as the contract's billing party, simplifying the purchasing process for agencies accustomed to standard procurement procedures. It's another step in the company's expansion beyond the commercial sector, following its earlier move to make Claude available on Amazon Bedrock for FedRAMP High and DoD IL4/IL5 workloads.

Market Significance

Anthropic's entry into US public administration is part of a broader race among AI vendors for government contracts, one where OpenAI and Microsoft are also competing with their own public-sector products. For tech companies, government contracts mean stable, multi-year revenue, but also require meeting far stricter security requirements than in the commercial segment.

For Polish companies and institutions watching the AI market, this signals the direction security standards for agentic tools used in government are heading. Similar requirements around auditability of AI agent actions and spending controls at the organizational-unit level will likely appear in European public tenders as the AI Act is implemented.

Sources: Bringing Claude Code and Claude Cowork to Government (claude.com), Anthropic Brings Claude Code and Cowork to Government (letsdatascience.com)