Compositional Learning-And-Reasoning for AI Complex Systems Engineering (CLARA) is sponsored by DARPA. The CLARA program seeks innovative research proposals to develop high-assurance machine learning/AI reasoning composition approaches, including theory, algorithms, and open-source software implementations. The program emphasizes verifiable AI systems and neurosymbolic architectures, with applications in course-of-action planning, multi-condition medical guidance, and supply chain logistics. All software developed must be open source under an Apache 2.0 license.

The DARPA CLARA program seeks to create high-assurance AI by tightly integrating machine learning with automated reasoning. Rather than the current industry approach of loosely coupling ML with reasoning as an afterthought, CLARA funds research into deep compositional integration that produces AI systems with strong logical explainability and computational tractability. The program targets applications in autonomous systems, command and control, kill web operations, supply chain logistics, wargaming, and medical, financial, and legal domains. TA1 funds development of new high-assurance ML/AR composition approaches including theory, algorithms, and open-source code. TA2 builds a software composition library that integrates validated TA1 tools into a common framework. All software deliverables must use permissive open-source licenses. The program is managed by Benjamin Grosof in DARPA's Defense Sciences Office. Solicitation DARPA-PA-25-07-02 was published February 10, 2026, with full proposals due April 17, 2026 (extended from April 10 via Amendment 1).

DARPA's CLARA (Compositional Learning-And-Reasoning for AI Complex Systems Engineering) program is a fundamental research initiative seeking to tightly integrate automated reasoning (AR) with machine learning (ML) to create high-assurance AI systems that are demonstrably trustworthy, not just empirically good. Published under solicitation DARPA-PA-25-07-02 on February 10, 2026, the program explicitly criticizes the current industry approach of tacking automated reasoning onto large language models and instead demands tight compositional integration of ML and AR components with hierarchical structure and transparent operation. CLARA is structured in two technical areas: TA1 funds development of new high-assurance ML/AR composition approaches including theory, algorithms, and open-source code, while TA2 builds a software composition library integrating validated TA1 tools into a common framework. All software must be released as open source under Apache 2.0 license. Application domains include autonomous systems, command and control, kill web operations, supply chain logistics, wargaming, and medical, financial, and legal domains. The program targets composition of Bayesian methods, neural networks, and logic programs. DARPA aims to execute awards by June 9, 2026, within 120 days of posting.