Two Years In: OpenMined Deep Partnerships under the NAIRR

This month, OpenMined is heading back to Arlington, VA, for the NAIRR’s second annual meeting. A lot has changed since last year’s inaugural gathering. The program is transitioning from a proof-of-concept to permanent national infrastructure, NSF has put $35 million on the table to establish a permanent operations center, and OpenMined just received approval to launch our first Deep Partnership project — a collaboration with Indiana University to build privacy-preserving tools for medical text de-identification. As one of the non-governmental partners in this initiative, we wanted to share what year two has looked like for us and what we see ahead.

From Pilot to Permanent: The NAIRR’s Maturation

The NAIRR Operations Center

In September 2025, NSF released a $35 million solicitation to establish the NAIRR Operations Center (NAIRR-OC), which will serve as the permanent organizational backbone for the program. The NAIRR-OC will handle day-to-day coordination, portal development, partner integration, and community building, transitioning these functions from the pilot infrastructure to a sustainable long-term model. This represents the single largest investment in making the NAIRR permanent.

Legislative Momentum

The CREATE AI Act (H.R. 2385) was reintroduced in March 2025 with bipartisan support from Representatives Obernolte (R-CA) and Beyer (D-VA). The bill would formally codify the NAIRR into law. Notably, the 2025 version includes a new provision allowing the program to accept donations from the private sector and federal agencies, reflecting the current fiscal environment. The Trump Administration’s AI Action Plan also explicitly calls for expanding the NAIRR, providing additional political tailwinds.

Growing the Ecosystem

Since its launch in 2025, the NAIRR consortium has grown to include 14 federal agencies and 28 private-sector partners.

NSF selected 10 new datasets for integration, spanning domains from lidar terrain mapping to microbiome data and software supply chain security. New program tracks, including Start-Up Project Request for quick-start projects and Deep Partnerships for sustained collaboration, have broadened how researchers can engage with the program.

What OpenMined Has Been Doing

Our year two engagement has been markedly more active than year one, spanning researcher support, deep partnership proposals, community education, and policy engagement.

Researcher Support

Our flagship year-one project, “Privacy-Preserving Machine Learning for Student Retention” at Concordia University Wisconsin, was completed late last year, achieving important milestones for research on cross-institution FERPA-protected data. You can read more about it here. The project demonstrated a public-private architecture for privacy-preserving research that is now serving as a reference for future NAIRR projects.

Launching Deep Partnerships

The introduction of the NAIRR Deep Partnership track in August 2025 was an exciting moment for OpenMined. This new track enables sustained, hands-on collaboration between NAIRR partners and research teams, hardening and formalizing the model we’d been testing and exploring with the NAIRR Pilot program staff.

We received multiple Deep Partnership proposals, and recently received approval from the Allocations Working Group to launch NAIRR250484: “Efficient Privacy-Preserving De-identification of Medical Text Using Distilled Language Models in Confidential Computing Environments.” Led by a professor from Indiana University, this project combines model distillation with trusted execution environments to build a lightweight, HIPAA-compliant pipeline for de-identifying protected health information in medical records.

We also reviewed proposals spanning biomarker discovery for Alzheimer’s disease, multi-modal AI-powered drug discovery for cancer treatment, and clinical diagnostic tools. While not every proposal was a fit, the volume of healthcare-focused requests validated our thesis that privacy-preserving AI is becoming a prerequisite for meaningful medical AI research.

Key Themes from Year Two

The Deep Partnership Model Works

Year one taught us that simply listing resources on a portal is not enough. Researchers need hands-on collaboration, training, and ongoing support. The Deep Partnership track addresses this by creating structured, sustained engagements between partners and research teams. For OpenMined, this model is ideal, as our tools require understanding a researcher’s specific data topology, institutional constraints, and collaboration needs.

State and Regional Hubs Are Emerging

A CRA-NSF Industry-Philanthropy Summit explored how state consortia could pool resources to provide for their communities. EmpireAI is a leading-example. This state-based approach could dramatically expand the NAIRR’s reach, particularly to under-resourced institutions that currently make up only a fraction of NAIRR users.

International Interest Is Growing

The UK launched its own AI Research Resource (AIRR) in July 2025, accepting proposals for supercomputing access. It joins a growing ecosystem of national and regional AI compute programmes: the EU’s expanding network of EuroHPC AI Factories (with 19 sites now selected across Europe), the IndiaAI Mission (building 18,000+ GPUs of public compute capacity), and Japan’s AI Bridging Cloud Infrastructure facility. While the operating models of these initiatives differ, their parallel proliferation creates opportunities for collaboration. OpenMined is exploring whether researchers across these programmes could benefit from leveraging our cross-institutional collaboration tools.

What We’ve Learned

1. Lower the barrier to entry. Many researchers who would benefit from our tools don’t have deep technical backgrounds in privacy-preserving AI. We’ve been drafting clearer descriptions and building more accessible onboarding materials so researchers can understand what they’re choosing and self-select into the right engagement.

2. The approval process matters. In year one, we encountered friction with proposals being rejected without clear recourse, delays in the review pipeline, and limited ability to engage with applicants before decisions were made. We fed these lessons back to the NAIRR team, and year two has seen significant improvements, including the Start-Up track’s faster approval cadence and the Deep Partnership track’s more collaborative model.

3. Healthcare is the proving ground. The overwhelming majority of proposals we receive involve healthcare data—medical records behind HIPAA restrictions, clinical notes, genomic data under sharing policies. If privacy-preserving AI can work in healthcare, it can work anywhere. Our collaboration with Indiana University is positioned to serve as a reference implementation for this thesis.

4. Organic growth compounds. Each supported researcher becomes a node in a growing network of privacy-preserving AI practitioners. We are excited to continue building this network!

Looking Ahead

We look forward to attending the NAIRR’s second annual meeting. With the program transitioning out of its pilot stage, $35 million in operational funding on the table, and bipartisan legislation introduced in Congress, the NAIRR is poised to become a defining feature of America’s AI research landscape.

For OpenMined, year three is about scaling what works. We’re focused on deepening our Indiana University collaboration into a publishable case study, expanding our Deep Partnership portfolio to include more multi-institutional consortia, building more accessible tools and documentation for researchers who are new to privacy-preserving AI, and engaging with the emerging state and regional hub model to ensure privacy-preserving resources are part of the conversation.

Get Involved

If you’re a researcher working with sensitive data, we want to hear from you. OpenMined offers free software for distributed, privacy-preserving data science, compute credits for research consortia, and training sessions to get your team up and running.

Whether you’re already part of the NAIRR or considering applying, we can help you unlock insights from data you couldn’t access before.

Get started: Visit nairrpilot.org to submit a proposal, or reach out to us directly through the OpenMined NAIRR page.