AI Startups Build Secure Backbone for US Intelligence

Little-known infrastructure firms solve the Pentagon's biggest AI challenge: using powerful language models without leaking classified data.

Apr. 11, 2026 at 2:08pm

Got story updates? Submit your updates here. ›

A highly detailed 3D illustration of a glowing, futuristic server rack filled with illuminated circuit boards, cables, and other digital infrastructure components, conceptually representing the secure backbone enabling the use of AI in sensitive government applications.Secure AI infrastructure powers the intelligence community's use of powerful language models without exposing classified data.Washington Today

A small group of AI infrastructure startups is building secure systems that allow the Pentagon and intelligence agencies to use large language models on classified data without exposing sensitive information. These firms provide the digital equivalent of hardened facilities, ensuring model inference happens entirely inside government-controlled environments with no data flowing back to the AI developer. This air-gapped or tightly controlled deployment approach is what the intelligence community requires before trusting any commercial model with classified material.

Why it matters

As the use of AI accelerates across industries, the demand for infrastructure that keeps proprietary data shielded from model training pipelines is surging. These startups are positioning themselves as the essential middleware of the AI economy, solving a critical challenge for both government agencies and large enterprises in sectors like healthcare, finance, and manufacturing.

The details

The core challenge is managing the tension between the utility of powerful language models and the risk of exposing sensitive data. Train a large language model on proprietary information, and it becomes a remarkably powerful assistant - but letting the wrong person query that model can lead to catastrophic leaks. For the intelligence community, the consequences of getting this wrong are measured in compromised operations and human lives, not just competitive disadvantage.

  • In 2024, Anthropic partnered with Palantir and Amazon Web Services to provide secure cloud hosting and software platforms for its Claude model, allowing it to be used on the Defense Department's classified networks.
  • Earlier this year, Anthropic clashed publicly with the Pentagon over ethical red lines around domestic surveillance and autonomous weapons, leading to a temporary bar on federal agencies and contractors working with the company.

The players

Nicolas Chaillan

The founder of AI infrastructure platform Ask Sage, who estimates the secure AI infrastructure market currently sits at roughly $2 billion and is expanding rapidly as defense teams demand tools that work inside existing security perimeters.

Emily Harding

A former CIA analyst now at the Center for Strategic and International Studies, who frames the challenge as a difficult balance - feed a model enough data and it knows too much, withhold enough data and it cannot do its job.

Anthropic

An AI company whose Claude model was one of only a handful of large language models approved for use on the Defense Department's classified networks, until a recent legal battle disrupted the arrangement.

Palantir

A technology company that partnered with Anthropic and Amazon Web Services in 2024 to provide secure cloud hosting and software platforms for Anthropic's Claude model to be used on the Defense Department's classified networks.

Amazon Web Services

A cloud computing subsidiary of Amazon that partnered with Anthropic and Palantir in 2024 to provide secure cloud hosting and software platforms for Anthropic's Claude model to be used on the Defense Department's classified networks.

Got photos? Submit your photos here. ›

What’s next

As the use of AI continues to accelerate across industries, the demand for secure infrastructure that can protect sensitive data is expected to grow rapidly. The market opportunity for the startups building this critical middleware is projected to expand significantly in the coming years.

The takeaway

The work of these little-known AI infrastructure startups is foundational to enabling the intelligence community and other organizations to leverage the power of large language models without compromising sensitive data. Their secure deployment platforms are essential for bridging the gap between the utility of AI and the need to protect proprietary information.