Disruption through Intelligent Strategies, Counter Options, and Resilient Defenses (DISCORD)

The Defense Advanced Research Projects Agency (DARPA) is soliciting innovative proposals under the Disruption through Intelligent Strategies, Counter Options, and Resilient Defenses (DISCORD) program, a 24-month initiative structured into two 12-month phases aimed at developing an AI-native system capable of rapidly generating diverse, explainable tactical options using live sensor data and high-fidelity simulation. The program seeks to enable commanders with a portfolio of adaptive strategies that can pivot dynamically in response to evolving battlefield conditions, ensuring strategic flexibility while preserving human judgment by producing outcomes that are transparent and verifiable. Core technical innovations required include abstraction of physics-based reduced order modeling, AI-driven simulation executed at the edge, online model-predictive control and dynamics learning, recurrence quantification analysis, integration with live sensor feeds, and application of game theory and behavioral theory to model adversary decision-making. DISCORD is part of Ender’s Foundry, a Department of War Pace-Setting Project designed to accelerate AI-enabled simulation capabilities to maintain a strategic advantage over AI-adapted adversaries. All work is classified as Controlled Unclassified Information (CUI), and access to detailed technical and evaluation requirements is restricted through a formally requested CUI addendum, obtainable via submission of a designated request form to DARPA. Performance is measured through a series of quantitative metrics across three technical areas, with success determined by progression through clearly defined milestones and formal evaluations. In Phase 1, performers must demonstrate a "steel thread" end-to-end workflow by Month 3, deliver a Minimum Viable Product simulation dry run and live hardware demonstration by Month 6, and achieve formal evaluation against Phase 1 benchmarks by Month 12, including targets such as commanding decision times under five minutes, intent translation accuracy above 80 percent, and detection of over 75 percent of pre-identified vulnerabilities. Down-selects occur at Months 3, 6, and 12 based on integration success and metric achievement. Phase 2 begins with a joint workshop at a government test facility, culminates in a fully integrated demonstration across TA 1, TA 2, and TA 3 systems in a joint UxS scenario by Month 18, and concludes with a scored "graduation" exercise in a full Live, Virtual, and Constructive (LVC) environment by Month 24, evaluated by MITRE using targets such as decision times under 60 seconds, intent translation accuracy above 95 percent, and vulnerability detection exceeding 98