Software testing corrects errors in software and provides assurance that software functions as intended

The National Institute of Standards and Technology awarded Purdue University $249,595 on September 1, 2026, under the Measurement and Engineering Research and Standards program (CFDA 11.609) to develop an automated software testing system powered by artificial intelligence.

Purdue will further develop an AI auditor capable of diagnosing bugs, generating test cases, and repairing software errors across five programming languages: C/C++, Java, Python, Ada, and Go. The project includes developing a subcomponent to extract formal specifications from code and conducting experiments to evaluate the AI auditor's quality in testing, bug detection, and error repair. Purdue will create and identify datasets including code and specifications for evaluating code testing and generation systems, with results intended for release to the research community and public through GitHub under a CC-BY-4.0 license.

Deliverables include a developed and usable AI auditor, a robust collection of code and specification data for use in related experiments, experiment reports evaluating the AI auditor's performance, and other grant deliverables. No subawards are planned.

Performance occurs in West Lafayette, Indiana, from September 1, 2026, through August 31, 2027.

Activities

The activities to be performed are:1. Further development of their AI auditor, which will diagnose, identify bugs, and generate test cases for software in five different programming languages: c/c++, java, python, ADA, and GO2. Development of a subcomponent that extracts formal specifications from CODE3. Experiments evaluating the quality of the AI auditor to test code, find bugs, and repair software BUGS4. Creation and identification of data to evaluate code testing and code generation systems. This data includes code and specificationsexpected outcomes: the expected outcomes are:1. A developed and usable AI auditor that can generate test cases, find errors in software, and repair software errors.2. A robust collection of data including code and specifications that can be used in other software code experiments including related experiments that NIST may run.3. Experiment reports that evaluate the quality of the AI auditor.4. Other grant deliverables.

Intended beneficiaries

The proposers will likely use their AI auditor as part of future work or in other proposals. Furthermore, as the proposers claim to release the BUG corpus and much of the data publicly (through github and under a CC-by-4.0 license), the research community and the public also plan to benefit.

Subrecipient activities

There are no planned subaward activities.