DARPA Launches CLARA Program for High-Assurance AI Systems
March 9, 2026 · 2 min read
Granted Research Team · Editorial policy
DARPA's Defense Sciences Office has opened solicitations for CLARA, a new fundamental research program aimed at building AI systems that can prove — not just claim — they work correctly.
The Compositional Learning-And-Reasoning for AI Complex Systems Engineering program, published on SAM.gov as solicitation DARPA-PA-25-07-02, targets one of AI's most stubborn problems: how to compose machine learning and automated reasoning components into systems that scale to real-world complexity while maintaining verifiable guarantees.
What DARPA Wants Built
CLARA defines assurance as "verifiability with strong explainability to humans, based on automated logical proofs and hierarchical, vetted logic building blocks." That's a significant bar. The program wants researchers to develop methods for tightly integrating Bayesian networks, neural networks, and logic programs into composite systems that can be mathematically verified.
Two technical areas structure the work. TA1 funds development of new composition approaches — the theory, algorithms, and open-source code for building high-assurance ML/AR hybrids. TA2 builds a software composition library that integrates validated TA1 tools into a common framework other researchers and defense developers can use.
Target domains include autonomous systems, command and control, supply chain logistics, medical applications, financial systems, and scientific design.
Funding and Timeline
Awards run up to $2 million over 24 months. DARPA aims to execute awards by June 9, 2026, giving selected teams a fast start. All software must be open-sourced under a commercialization-friendly license, with Apache 2.0 preferred.
Proposals are due April 10, 2026, at 4:00 p.m. Eastern. Program manager Benjamin Grosof leads the initiative; questions can be directed to [email protected].
Why This Matters for AI Researchers
CLARA sits under DARPA's broader AI Forward initiative, which received $310 million in the FY2025 budget. AI Forward's organizing principle is trustworthiness — building AI that works reliably in adversarial, ambiguous, and high-stakes environments. CLARA operationalizes that principle with a specific technical agenda.
For AI researchers working at the intersection of formal methods and machine learning, this is one of the few funded programs demanding both mathematical rigor and practical scalability. The April 10 deadline leaves roughly four weeks. In-depth analysis of DARPA's full AI program portfolio is available on the Granted blog.