Accelerating the time to science for the CDC and NIH

The AI transition from experimental chatbots to agents faces a critical hurdle in highly-regulated sectors like biomedicine and public health. While these autonomous systems can navigate complex data to perform real-world tasks, there is a clear lack of a security-focused,  standardized, and observable framework for AI agents to interact with sensitive data. Without enterprise-grade guardrails, the potential of agentic workflows remains confined to the lab sandbox.

To address this challenge, Red Hat has launched a joint research effort with the Centers for Disease Control and Prevention (CDC) and the National Institutes of Health (NIH). By hardening the open source Model Context Protocol (MCP), we are establishing the infrastructure requirements necessary to more safely move agentic workflows into production environments where accuracy and data security are non-negotiable.

Bridging the gap between AI and sensitive repositories

In the context of public health, the "time to science" - the interval between a research question and a data-driven answer - is a matter of national security. Fragmented data silos and poorly documented API calls currently hamper this process, with workflows taking hours. By applying the Model Context Protocol to these complex environments, that window is compressed to just seconds.

This co-engineering work addresses the four primary barriers preventing researchers working with AI agents from making meaningful scientific impact:

• Replacing latency with instant access: By creating a standardized interface between AI models and impactful public health repositories researchers can move from manual data gathering to real-time analysis.
• Contextual integrity over raw data: In scientific research, a conclusion  without context  is incomplete. The current focus includes expanding MCP to confirm that AI agents do not just receive data, but also the critical errata, caveats, and footnotes necessary to prevent the spread of misinformation.
• Auditability as a requirement: To move from experimental assistants to autonomous agents, every interaction must be traceable. This effort establishes the governance and observability patterns required to make AI-driven decisions fully auditable and compliant with federal policy.
• Knowledge acceptance testing: When applications are deployed in a regulated industry, they undergo strenuous testing and acceptance by a domain expert who is responsible for them. MCP does not currently have a method for Acceptance testing, and our co-engineering effort directly addresses this gap using novel extrapolation at scale.

From agency collaboration to OpenShift AI innovation

While MCP is already a core component of Red Hat OpenShift AI, the insights and technical hardening from our work with the CDC and NIH serve as a powerful feedback loop. This collaboration will help directly shape the MCP upstream community and by extension the next generation of OpenShift AI capabilities.

By treating these public health use cases as the ultimate stress test, we are refining how OpenShift AI manages MCP servers, handles high-concurrency data requests, and drives security-enhanced connections between models and diverse data sources. These improvements - ranging from better scaling logic to more robust error handling for complex APIs - will ultimately flow back into the OpenShift AI platform.

This means that every Red Hat customer, regardless of industry, can benefit from a platform battle-tested by the world's leading scientific institutions.

Open standards for a global impact

The goals of this joint research effort extend beyond immediate technical gains; they can help shape the global approach to agentic AI. By open sourcing the testing frameworks developed alongside the federal partners, we are helping the broader healthcare and science community overcome the cultural and organizational hurdles of moving to an agentic framework.

As enterprises move toward autonomous AI agents, we want our collaboration with the CDC/NIH to establish the infrastructure to make these agents faster, safer, and - most importantly - trusted by the people who rely on them to protect public health.