A Survey of Quality Risks and Software Assurance Techniques for AI-Generated Code

PhD Qualifying Examination


Title: "A Survey of Quality Risks and Software Assurance Techniques for 
AI-Generated Code"

by

Mr. Chenyang SUN


Abstract:

AI coding is shifting software development from human-led implementation 
toward agentic workflows that interpret issues, modify repositories, run 
tests, and repair patches. Although this shift makes code generation faster 
and cheaper, it widens the gap between code production and human 
understanding. Developers now review changes they neither designed nor wrote, 
often produced faster than they can assess their behavior and repository 
effects. Accepting such changes therefore depends on software-assurance 
evidence.

This survey examines AI-generated code from two complementary perspectives. 
First, it synthesizes empirical evidence on quality risks and their causes. 
Second, it surveys static analysis, dynamic testing, formal verification and 
specification inference for assessing AI-generated code. These techniques 
support assurance claims with different scopes and strengths rather than a 
uniform correctness guarantee.

We argue that quality claims about AI-generated code should rely on 
reproducible evidence whose scope matches the relevant risk and generated 
change, rather than on plausible output, model confidence, or limited test 
success.


Date:                   Wednesday, 23 September 2026

Time:                   4:00pm - 6:00pm

Venue:                  Room 3494
                        Lift 25/26

Committee Members:      Prof. Charles Zhang (Supervisor)
                        Prof. Qiong Luo (Chairperson)
                        Dr. Dimitris Papadopoulos