Generative AI raises short-term productivity by completing tasks learners would otherwise practice on their own. Whether this exchange erodes frontier skill depends on the mode of use: substitute-users let AI stand in for practice and fail to develop skills, while complement-users use AI to learn faster. The modes look alike in AI-aided output, so organizations screening on that output cannot tell them apart. We ask whether the AI-prohibited evaluation gates organizations already operate can separate the modes. In elite competitive programming, ICPC and IOI contests prohibit AI under in-person proctoring, with qualification-round entry, whereas Codeforces (CF) practice and contests are unproctored and open to all. From CF practice histories we build an AI-prompt signature consistent with AI usage, more first-attempt acceptances, fewer attempts and debugging retries. CF practice has shifted toward this signature across entry cohorts spanning two AI rollouts. On CF, a stronger signature predicts smaller rating gains for users with no ICPC-IOI affiliation, but not for those who qualified. Inside the AI-prohibited ICPC environment, AI-era entrants show no skill erosion, and shifts toward AI-style practice predict higher non-AI-aided scores. One screening mechanism fits both: where the modes mix, a stronger signature flags substitute-users; among those who qualified, a strengthening signature marks adoption of the complement mode. The message is constructive: AI-style practice is compatible with frontier skill; the erosion risk links to the substitute mode; and separating the modes is a design question for the exams organizations regularly administer, from medical and legal boards to professional certification.

翻译：暂无翻译