ARLINGTON, Va. - The Defense Advanced Research Projects Agency (DARPA) is seeking proposals for high-assurance artificial intelligence under a new Disruption Opportunity (DO) issued by its Defense Sciences Office as part of the Disruptioneering program (DARPA-PA-25-07).
The solicitation, known as the CLARA program, calls for basic and applied research to develop artificial intelligence (AI) "systems of systems" that combine machine learning (ML) and automated reasoning (AR) with strong verifiability and explainability. Awards will be issued as Other Transaction (OT) prototype agreements with a total value of up to $2,000,000 per project, covering a 15-month Phase 1 base and a 9-month Phase 2 option.
DARPA is aiming to overcome a longstanding trade-off in artificial intelligence: ML techniques such as neural networks offer broad applicability but limited explainability, while AR approaches provide verifiability and logical transparency but struggle with scalability and flexibility. The CLARA effort seeks to integrate these approaches into a unified, theory-driven architecture that delivers both scalability and high assurance.
Related: Rethinking military batteries: researchers eye non-rigid energy storage for uncrewed and wearable systems
The program emphasizes the development of AI architectures that hierarchically compose ML and AR subsystems, enabling outputs from multiple models to be combined into a coherent system response. Proposed approaches must demonstrate strong logical explainability, computational tractability, and the ability to scale across large, complex problem sets.
CLARA tech
CLARA is structured into two technical areas. Technical Area 1 focuses on developing new high-assurance ML/AR approaches, including theory, algorithms, and open-source software for inference and training. Performers will also be required to demonstrate adaptability to new data and scalability to large problem sizes. Technical Area 2 centers on building a software integration library to enable interoperability among Technical Area 1 solutions, including common data formats, interfaces, and end-to-end explainability tools.
The program will include coordinated activities such as workshops and hackathons to evaluate interoperability and performance across teams. Example application domains include course-of-action planning, multi-condition medical decision-making, and supply chain optimization, all of which require combining predictive analytics with rule-based reasoning.
Phase 1 will focus on initial theory development, experimentation, and benchmarking against state-of-the-art systems, while Phase 2 will extend these approaches to additional AI methods and demonstrate more advanced system integration and training capabilities. Milestone-based payments will be tied to demonstrations, reports, and participation in program-wide events, including competitive hackathons.
DARPA expects submissions to be unclassified and to address either Technical Area 1 or Technical Area 2. Proposals are due by 17 April 2026 at 4:00 p.m. EDT, and the agency aims to complete awards within 120 days of the announcement. DARPA asks intersted parties to contact them via email at [email protected].
For more information, please visit https://sam.gov/workspace/contract/opp/ed634c7caae44ea19bfbefef912862f4/view.
