Anthropic launches Claude Gov for US classified intelligence operations

Anthropic has launched Claude Gov, specialized AI models designed for US national security agencies to handle classified information and intelligence operations. The models are already serving government clients in classified environments, marking a significant expansion of AI into sensitive national security work where accuracy and security are paramount.

What you should know: Claude Gov differs substantially from Anthropic’s consumer offerings, with specific modifications for government use.
• The models can handle classified material and “refuse less” when engaging with sensitive information, removing safety restrictions that might block legitimate government operations.
• They feature “enhanced proficiency” in languages and dialects critical to national security operations.
• Access is restricted exclusively to personnel working in classified environments.

How it works: The specialized models support various intelligence and defense functions across government agencies.
• Claude Gov handles strategic planning, intelligence analysis, and operational support for US national security customers.
• The models are customized specifically to process intelligence and defense documents.
• Anthropic says the new models underwent the same “safety testing” as all Claude models.

The competitive landscape: Major AI companies are increasingly competing for lucrative government defense contracts.
• Microsoft launched an isolated version of OpenAI’s GPT-4 for the US intelligence community in 2024, operating on a government-only network without internet access and serving about 10,000 individuals.
• OpenAI is working to build closer ties with the US Defense Department, while Meta recently made its Llama models available to defense partners.
• Google is developing a version of its Gemini AI model for classified environments, and Cohere is collaborating with Palantir for government deployment.

Why this matters: The push into defense work represents a notable shift for AI companies that previously avoided military applications.
• Anthropic has been pursuing government contracts as it seeks reliable revenue sources, partnering with Palantir and Amazon Web Services in November to sell AI tools to defense customers.
• However, using AI models for intelligence analysis raises concerns about confabulation, where models generate plausible-sounding but inaccurate information based on statistical probabilities rather than factual databases.
• These risks are particularly critical when accuracy is essential for national security decisions, as AI models may produce convincing but incorrect summaries or analyses of sensitive data.