Bid document

Special Notice
Request for Information
Formal Learning and
Abstractions for System
Heterogeneity (FLASH)

DARPA-SN-27-17

November 3, 2026

Defense Advanced Research Projects Agency
Information Processing Techniques Office
675 North Randolph Street
Arlington, VA 22203-2114

DARPA-SN-27-17

REQUEST FOR INFORMATION (RFI) DESCRIPTION

The Defense Advanced Research Projects Agency (DARPA) Information Processing Techniques Office (IPTO) seeks information on research related to Formal Learning and Abstractions for System Heterogeneity (FLASH). Specifically, this RFI seeks information on (i) national security applications that could benefit from increased robustness and other operational properties enabled by heterogeneous engineering solutions; and (ii) potential techniques, informed by current research and practice, for the automated synthesis and understanding of high-assurance cyber-physical engineering systems from formal specifications.

RFI BACKGROUND

DARPA is exploring novel methodologies that augment generative-model-based agents with symbolic reasoning techniques to formally represent design intent, and to rapidly and efficiently synthesize semantically heterogeneous cyber-physical engineering systems together with design and assurance collateral to aid system understanding. By systematically decoupling high-level system requirements from instance-level design details, the FLASH framework is intended to operationalize robustness through heterogeneity. Under this paradigm, generated instances may exhibit meaningful variation in failure modes and attack surfaces while preserving specified system properties. Exploitation of such variation could reduce the durability and transferability of instance-specific adversary knowledge, including knowledge of discovered vulnerabilities, configurations, or common fault modes. Co-generation of design collateral for assurance and understanding helps provide evidence that correctness, heterogeneity, and resilience can be formally assessed rather than evaluated solely empirically. Key technical thrusts in FLASH include (i) formalizing system requirements, operational intent, and semantic invariants while exposing valid design degrees of freedom that allow for heterogeneous alternative choices; (ii) developing models and techniques to generate different system design instances while guaranteeing functional equivalence and formally provable correctness; and (iii) techniques to co-generate assurance and explanation artifacts that are