Arlington, Virginia – U.S. military researchers are calling on industry to increase the assurance levels and expansion capabilities of artificial intelligence (AI) systems.
Officials at the U.S. Defense Advanced Research Projects Agency (DARPA) in Arlington, Virginia, have issued a call for applications (DARPA-PA-25-07-02) for the Constructive Learning and Reasoning for AI Complex Systems Engineering (CLARA) program.
CLARA aims to create reliable and widely applicable AI systems that include machine learning and automated reasoning (also known as knowledge representation and reasoning subsystems).
Contractors are expected to combine interoperable integration of higher-order logic, probabilistic logic, logical expressiveness, hierarchically structured knowledge representation, automated reasoning and machine learning, including neural networks, Bayesian machine learning, reinforcement learning, generalized additive models, logic programs, classical logic, and answer set programs.
The need for highly reliable AI
The military’s need for high-assurance AI has slowed its adoption, and DARPA researchers explain that the trade-off between machine learning and automated inference is holding back high-assurance. Furthermore, machine learning is difficult to explain and therefore has weak guarantees.
Although there are several examples of systems that offer preliminary promise that automated inference-based machine learning systems can be built, they have significant limitations and applicability and tractability challenges have yet to be overcome.
CLARA has two technology areas. One is how to develop reliable machine learning and automated inference. Development of software configuration libraries. It is not possible to select organizations in both technical areas.
A reliable machine learning and automated reasoning approach develops a reliable machine learning and automated reasoning theory. This work involves algorithms. Expressive and syntactic generalizations and extensions, specializations and restrictions, and corresponding characterization and verifiability guarantees. Computational tractability and scalability. Transforming between machine learning and automated inference.
open source software
Participants will create the first open source software to demonstrate approaches such as automated inference-based machine learning inference and training.
Participants will also tackle automatic inference-based machine learning training for large-scale problems, demonstrating the scalability of this training and confirming that automatic inference-based machine learning models can quickly adapt to previously unseen data with minimal additions or manual human editing of training data.
Interested companies must submit responses to the DARPA BAA tool online at https://baa.darpa.mil by April 10, 2026.
For questions or concerns, please email Benjamin Grosof, DARPA CLARA Program Manager. [email protected]. For more information, please visit us online at https://sam.gov/workspace/contract/opp/3530b2c0a68d4de786079e7305d4f625/view.
