DARPA launches CLARA high-assurance AI program

The Defense Advanced Research Projects Agency (DARPA), the Pentagon’s research and technology arm, issued a solicitation for its Compositional Learning-And-Reasoning for AI Complex Systems Engineering (CLARA) program, inviting proposals for high-assurance artificial intelligence research.

The solicitation was released by DARPA’s Defense Sciences Office and seeks innovative basic or applied research concepts in “high assurance artificial intelligence systems”. Proposals are due by 4:00 p.m. Eastern Time on April 10, 2026, with abstracts encouraged by March 2, 2026.

According to the announcement, CLARA is “an exploratory, fundamental research program that aims to create high assurance, broadly applicable AI systems of systems”. The program will pursue a “scientific, theory-driven architectural foundation for the hierarchical composition” of Machine Learning (ML) and Automated Reasoning (AR) subsystems.

DARPA states that assurance under CLARA “means verifiability with strong explainability to humans, based on automated logical proofs and hierarchical, vetted logic building blocks”. The agency anticipates performers will combine approaches such as higher-order logic, probabilistic logic, and interoperable integration of AR and ML.

The program will be structured in two technical areas. Technical Area 1 (TA1) focuses on developing new high-assurance ML/AR composition approaches, including theory, algorithms, and open-source software implementations. Technical Area 2 (TA2) will create a software composition library to integrate and support validated TA1 tools into a common framework.

As detailed in the solicitation, awards will be made as Other Transactions for prototype projects, with a total award value for combined Phase 1 and Phase 2 efforts limited to $2,000,000. Phase 1 will run for 15 months and Phase 2 for 9 months, for a maximum 24-month period of performance.

The program sets specific performance metrics, including “Verifiability without loss of performance,” composition of multiple AI kinds, and “Computational Time complexity is Polynomial”. For TA1 performers, Phase 2 also introduces sample complexity requirements in adapting models to new tasks.

CLARA will include program-wide activities such as workshops and hackathons. According to the description, hackathons will involve wide-scope integration scenarios developed with an Independent Verification and Validation team, with examples including “a partial kill web integrating limited components of target recognition, tracking, weapons selection, triaging decision support”.

Software developed during the program is expected to be open sourced with a commercialization-friendly license, preferably Apache 2.0. The agency’s goal is to execute awards within 120 calendar days of the posting date, with a target of June 9, 2026, for award execution.

DARPA’s solicitation outlines potential application domains for CLARA technologies, including course-of-action planning, multi-condition medical guidance, and supply chain and logistics problems.