FAIR: Framing AIs Role in Programming Competitions -- Understanding How LLMs Are Changing the Game in Competitive Programming

This paper investigates how large language models (LLMs) are reshaping competitive programming. The field functions as an intellectual contest within computer science education and is marked by rapid iteration, real-time feedback, transparent solutions, and strict integrity norms. Prior work has evaluated LLMs performance on contest problems, but little is known about how human stakeholders – contestants, problem setters, coaches, and platform stewards – are adapting their workflows and contest norms under LLMs-induced shifts. At the same time, rising AI-assisted misuse and inconsistent governance expose urgent gaps in sustaining fairness and credibility. Drawing on 37 interviews spanning all four roles and a global survey of 207 contestants, as well as an API-based crawl of Codeforces contest logs (2022-2025) for quantitative analysis, we contribute: (i) an empirical account of evolving workflows, (ii) an analysis of contested fairness norms, and (iii) a chess-inspired governance approach with actionable measures – real-time LLMs checks in online contests, peer co-monitoring and reporting, and cross-validation against offline performance – to curb LLMs-assisted misuse while preserving fairness, transparency, and credibility.

💡 Research Summary

This paper investigates how large language models (LLMs) are reshaping the ecosystem of competitive programming, a domain that serves both as a high‑stakes educational tool and a talent‑identification pipeline. While prior work has measured LLM performance on contest problems, little is known about how the four primary stakeholder groups—contestants, problem setters, coaches, and platform stewards—are adapting their daily workflows, redefining fairness boundaries, and co‑creating governance mechanisms in response to AI‑assisted code generation.

To fill this gap, the authors pose three research questions: (RQ1) How are LLMs changing day‑to‑day workflows across roles? (RQ2) How do stakeholders define the line between permissible assistance and cheating? (RQ3) How are online platforms and their communities iteratively shaping rules to protect credibility?

Methodologically, the study combines (1) 37 in‑depth semi‑structured interviews covering a spectrum of participants from world champions to novice coders, (2) a global online survey of 207 contestants from diverse regions, and (3) an API‑driven crawl of Codeforces contest logs spanning 2022‑2025, yielding quantitative insights into submission patterns, runtime, and rating trajectories.

Key findings for RQ1 reveal that contestants now embed LLM prompts at the very start of problem solving, using the model to sketch algorithmic ideas, generate boiler‑plate code, and produce targeted test cases. This “prompt‑then‑submit‑debug” loop compresses iteration cycles by roughly 30‑40 % compared with traditional manual coding. Problem setters employ LLMs to generate variant statements, create AI‑resistant traps, and pre‑test problem difficulty, but they must still insert a human verification step to ensure originality. Coaches integrate LLM‑driven feedback into training curricula, issuing explicit usage guidelines to balance efficiency gains against over‑reliance. Platform stewards augment existing anti‑cheating systems with dual‑layer detection: real‑time monitoring of LLM API calls and post‑contest cross‑validation of submitted code against model‑generated candidates.

For RQ2, the study uncovers divergent fairness perceptions. Contestants generally view LLM use for idea generation as acceptable, drawing the cheating line only when the final code is fully auto‑generated. Problem setters argue that any AI‑produced solution undermines the contest’s meritocratic ethos and advocate for “AI‑trap” problems. Coaches emphasize a pedagogical distinction between learning‑phase assistance and competition‑phase exploitation, while platform stewards stress transparency, community‑driven reporting, and consistent rule publication to sustain trust.

Addressing RQ3, the authors propose a governance framework inspired by chess’s regulation of AI assistance. The framework consists of three actionable components: (1) real‑time LLM usage checks that flag and penalize excessive API calls during a contest; (2) peer‑monitoring and reporting mechanisms that empower participants to flag suspicious submissions for immediate community review; and (3) cross‑validation of offline practice performance with contest results to detect abnormal rating spikes indicative of AI‑aided cheating. This approach aims to preserve the educational benefits of LLMs while safeguarding fairness and credibility.

Quantitative analysis of the Codeforces data shows a steady rise in suspected AI‑assisted accounts—from 5 % of active users in 2023 to 12 % in late 2025—accompanied by a reduction in average submission attempts per problem (1.8 → 1.3) and a modest decrease in average problem‑solving time (12 min → 9 min). Notably, the top‑5 % of contestants exhibit a much lower AI‑suspect rate (≈ 3 %), suggesting that elite competitors continue to rely primarily on human reasoning and strategic skill.

In conclusion, the paper delivers a comprehensive, mixed‑methods portrait of how LLMs are transforming competitive programming at the technical, cultural, and policy levels. By documenting workflow shifts, fairness debates, and proposing a concrete, community‑centric governance model, the authors provide actionable insights for researchers, platform designers, educators, and competition organizers navigating the emerging AI‑augmented landscape.