Anthropic expands US government AI deal despite Pentagon blacklist

Expanding government access

Anthropic has announced a new agreement with the US General Services Administration (GSA) to broaden the reach of its Claude AI products. The deal allows federal, state, local, tribal, and territorial government organizations to access Claude Enterprise, Claude for Government, and the Claude API. This expansion ensures that a wide variety of public entities can integrate the company's technology into their daily operations.

The new arrangement provides eligible organizations with a 40% discount on token-based usage. A token is a unit of text that the AI processes to generate a response. By moving to this model, the government can access the tools at a lower cost while only paying for the actual amount of data they process. Anthropic estimates that this specific agreement could save the US government $1 billion in total costs.

Agencies under this deal will not be required to commit to a minimum spending amount. Instead, they will pay based on their specific usage requirements. This flexibility is designed to make it easier for various levels of government to adopt the technology without a large upfront financial commitment. The rollout will occur through direct access from Anthropic, via resellers, and eventually through major cloud platforms.

Growth in public sector use

The adoption of Claude AI within the United States government has seen significant momentum recently. Anthropic reports that approximately 500 government organizations at the federal, state, and local levels are currently using the platform. The company noted that the number of people using Claude for Government on a weekly basis has multiplied by about eight since February 2026.

Growth has been even more pronounced within federal civilian agencies, which have seen usage grow about 20-fold in just one year. Anthropic points to specific use cases to demonstrate the utility of the tool in government settings. For example, a month-long test with 34 developers from 10 different Utah state agencies showed that 71% of the group saved at least 20 hours of work.

The company also highlighted a scenario where a state government team used Claude to respond to a ransomware attack. According to Anthropic, the team was able to respond in eight minutes, whereas the typical response time for such an incident is between two and three days. While these specific figures and examples have not been independently verified, they illustrate the speed and efficiency the company aims to provide to public officials.

The Pentagon legal dispute

Despite the expansion of its government business, Anthropic remains in a complicated relationship with the US administration. The company is currently involved in a legal battle with the Pentagon over military AI safeguards. In February 2026, Anthropic declined to remove specific limits designed to stop its technology from being used for fully autonomous weapons or mass domestic surveillance.

Following that refusal, the Trump administration directed federal agencies to stop using Anthropic's technology. This led the Pentagon to designate the company a national security supply-chain risk in March. A national security supply-chain risk is a designation indicating that a company's products or services could pose a threat to the country's security infrastructure.

The legal situation has seen various turns in the court system over the last year. A California federal court struck down a government action in August, ruling it was an illegal form of retaliation. However, a US appeals court upheld the Pentagon's risk designation on September 25. Anthropic has sought a rehearing on the matter, but no new ruling has been identified in the latest reporting.

Shifting pricing models

The new GSA agreement represents a shift in how the government pays for AI services compared to previous offers. Previously, Anthropic provided a deal that gave eligible federal agencies access to its tools for a nominal price of $1 per user. That specific offer is currently scheduled to expire on October 31, 2026.

The new agreement moves away from that per-user model toward consumption-based billing. This means that instead of a flat fee for every employee who has an account, the government will pay according to the volume of AI services they actually consume. This shift allows for more precise budgeting for agencies that may only need the AI for specific tasks rather than constant, high-volume use.

This new pricing structure also includes authorized environments specifically for sensitive but unclassified work. By providing these environments, Anthropic aims to give government agencies a secure way to handle data that is not for public release but does not require the highest level of military classification. The agreement provides a lower-cost route for adoption even while the larger legal questions regarding military use remain unanswered.