Anthropic Expands Claude Access for Government Agencies

Anthropic has expanded the availability of Claude 3 Haiku and Claude 3 Sonnet to government users via the AWS Marketplace for the US Intelligence Community (IC) and AWS GovCloud. This move enables government agencies to leverage AI for citizen services, policymaking, and intelligence analysis within secure environments that meet stringent government security standards.

Secure Deployment via AWS

Claude 3 Haiku and Claude 3 Sonnet are now accessible through AWS GovCloud and the AWS Marketplace for the US Intelligence Community. By leveraging Amazon Web Services (AWS) infrastructure, Anthropic provides government agencies with the flexibility and security required for public sector operations.

Government-Specific Use Cases and Applications

Government agencies can utilize Claude for a variety of immediate and future applications to improve public service and national security:

Current Applications: Improving citizen services, streamlining document preparation and review, and enhancing policymaking through data-driven insights.
Future Potential: Creating realistic training scenarios, coordinating disaster response, optimizing energy grids for sustainability, and enhancing public health initiatives.

Adapted Service Agreements and Usage Policy

Anthropic has modified its service agreements and usage policies to accommodate the legal authorities and specific missions of government users.

Contractual Exceptions for Intelligence Analysis

Anthropic has implemented specific contractual exceptions to its general Usage Policy for selected government agencies. These exceptions permit the use of Claude for legally authorized foreign intelligence analysis, including:

Combating human trafficking.
Identifying covert influence or sabotage campaigns.
Providing advance warning of potential military activities to enable diplomatic prevention or deterrence.

Maintained Restrictions

Despite these exceptions, Anthropic maintains strict prohibitions against the use of its models for:

Disinformation campaigns.
The design or use of weapons.
Censorship.
Malicious cyber operations.

Safety Frameworks and Governance

Access to these government-specific policies currently applies only to models designated as AI Safety Level 2 (ASL-2) under Anthropic's Responsible Scaling Policy (RSP). Anthropic commits to the regular evaluation of these partnerships and their impacts to ensure AI serves the public interest while mitigating risks.

Collaboration with AI Safety Institutes

Anthropic continues to work with government bodies to establish effective testing and measurement regimes. This includes providing pre-release access of Claude 3.5 Sonnet to the UK Artificial Intelligence Safety Institute (UK AISI), which shared its pre-deployment testing results with the US AI Safety Institute (US AISI).