Deep Learning Translation of Unsafe Source and Binary into Safe Software

By virtue of a Phase I Defense Advanced Research Projects Agency (DARPA) Small Business Innovative Research (SBIR) contract, Code 13 Security (doing business as Dark Sky Technology) has developed a Technical Readiness Level (TRL) 5 technology to translate type-unsafe source and binary software into the type-safe language Rust. The design of this system is innovative in its use of machine learning and a context-aware intermediate representation (IR).   We aim to productize this Phase I capability into Bulletproof Code®, a TRL 9 product to abstract and synthesize software binaries and source code across programming languages. Although the benefit of being able to cost-effectively translate type-unsafe languages C/C++ into the type-safe language Rust is formidable, the product being proposed goes way beyond just translation. The Common Objective Representation (COR) being introduced in this Phase II effort will enable the synthesis process to support additional objectives, including the ability to correct programming errors in repositories and verify the results using formal assurance evidence.   We believe that once users verify that automated code remediation, to address either nefarious contributors or poor coding, is possible, they will be eager to implement such tools in their development environments. Having deployed tooling into software development pipelines in the past, Dark Sky Technology will be ready to assist in making code remediation a seamless addition to their processes.