# AIChilles: Automatically Uncovering Hidden Weaknesses in AI-Evolved Systems

Source: [AIChilles: Automatically Uncovering Hidden Weaknesses in AI-Evolved Systems](<https://www.sota2.com/research/paper/6a30d8a5d33dfb23e783ccea>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-06-26T05:00:12.659Z

Authors: Yajie Zhou, Ao Li, Ashwin Silla, Zaoxing Liu, Vyas Sekar

Published: 2026-06-14T14:24:25.000Z

## Abstract

The computer systems community has recently seen growing interest in AI-driven system evolution, where AI agents iteratively rewrite systems. Frameworks such as AdaEvolve and Engram report 12-60% score improvements over human-designed algorithms. While these results are promising, there are practical concerns if these AI-evolved programs can perform worse on unseen workloads and exhibit scalability regressions. Given the speed and scale of AI-generated code, we need automated mechanisms to uncover such identify hidden weaknesses in AI-evolved systems programs. To this end, we develop AIChilles that takes as input a baseline program $P$ and an AI-evolved program $P'$, AIChilles searches for valid workloads where $P'$ regresses relative to $P$ in correctness, runtime, memory usage, or output quality. To tackle the diversity in system applications, weakness types and potential bugs, AIChilles combines deterministic workload-parameter extraction, agent-based constraint inference, differential oracles, and code-frequency coverage to discover diverse failures. Across five system applications and 30 AI-evolved programs, AIChilles finds 49 distinct hidden weaknesses. We also show that explicitly including AIChilles in the AI-driven development lifecycle can mitigate several of these weaknesses.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Program Evolution on EPLB](<https://www.sota2.com/research/sota/program-evolution-on-eplb.md>): Program Evolution; EPLB; Score: 0.153.

- [Program Evolution on Cloudcast](<https://www.sota2.com/research/sota/program-evolution-on-cloudcast.md>): Program Evolution; Cloudcast; Score: 627.1.

- [Program Evolution on LLM-SQL](<https://www.sota2.com/research/sota/program-evolution-on-llm-sql.md>): Program Evolution; LLM-SQL; Score: 0.712.

- [Program Evolution on TXN](<https://www.sota2.com/research/sota/program-evolution-on-txn.md>): Program Evolution; TXN; Score: 2832.

- [Program Evolution on Prism](<https://www.sota2.com/research/sota/program-evolution-on-prism.md>): Program Evolution; PRISM; Score: 21.89.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2606.15834>)

[PDF](<https://arxiv.org/pdf/2606.15834.pdf>)