This Quick Look provides a short preview of the full session presented at RSAC™ 2025 Conference, now available on demand. This presentation grounds expectations of using LLM-based systems for fixing software and will present results of evaluating 100s of examples of insecure code drawn from real systems by various models from OpenAI. The results illustrate the strengths, weaknesses, progress, and gaps of different systems as they have evolved over recent years. Methods for closing the gaps will be discussed.

See the full session here.