On June 1, 2026, the Defense Advanced Research Projects Agency and the National Science Foundation formally launched AI Forge, a joint initiative designed to accelerate high-impact artificial-intelligence research for national-security use cases that commercial markets have so far neglected.
The program asserts a gap that policymakers and researchers have long worried about: while frontier AI companies race to build ever-larger models, the most consequential questions for defense and intelligence — questions about how those systems behave under stress, whether they can be reliably controlled, and how they hold up under deliberate attack — remain underexplored. Those problems lack immediate commercial applications, and the private sector has had little incentive to invest in them.
AI Forge establishes three research thrusts. The first, AI Interpretability, aims to make model behavior understandable in high-stakes settings rather than only in routine use cases. The second, AI Control, focuses on producing strong, verifiable evidence that model behavior remains bounded and auditable — work that becomes more urgent as advanced systems grow more autonomous. The third, Adversarial Robustness, seeks to build AI that preserves its intended performance even when actively attacked by an adversary.
The program is designed to run for multiple years and to revisit its identified research challenges every six months to keep pace with the rapidly changing technical landscape. A steering body — composed of university representatives, frontier AI companies, and officials from more than 15 Defense Department and Intelligence Community agencies — will guide funding priorities.
The initiative's first concrete action is a Request for Information directed at the U.S. university research community, with a submission deadline of today, June 22, 2026. The RFI asks universities to identify capabilities that address the program's 15 stated research challenges and to indicate interest in participating in the initiative's formal research forum, which is scheduled to launch in summer 2026.
DARPA emphasized that the forum is intended to bridge a structural disconnect in AI research. Universities provide foundational science and train the next generation of AI talent. Frontier companies provide computing scale and production models. High-risk, high-reward work requiring both is difficult to pursue in either environment alone. By formally connecting the two — with mission-driven input from national-security agencies embedded throughout — AI Forge tries to address that gap.
The program is aligned with the White House's America's AI Action Plan and marks the federal government's most explicit bid to keep U.S. national-security AI research competitive at a time when China is investing heavily in dual-use AI capabilities. Researchers caution, however, that the pace of foundational commercial AI progress typically outpaces government programs' timelines. Whether AI Forge can operate with sufficient agility to remain relevant will depend heavily on how quickly its steering body can adapt priorities as the state of the art shifts.
Image source: v3b.fal.media