作弊
心理学
计算机科学
社会心理学
人工智能
可靠性(半导体)
考试(生物学)
计算机安全
面子(社会学概念)
应用心理学
作者
Tori Bakeman,Louis Hickman
标识
DOI:10.5465/amproc.2026.20793poster
摘要
The emergence of large language models (LLMs), such as those accessed through ChatGPT, has raised new concerns for unproctored, high-stakes assessments. Applicants can easily use LLMs to obtain external assistance, potentially undermining test validity. Drawing on ability-motivation-opportunity theories of faking, we examine how integrity warnings and webcam-based facial validation influence test-taker behavior and performance in the LLM era. Using a large field sample of job applicants completing a cognitive ability test for selection, we test whether cheating-related behaviors and test performance changed following the release of chat-based LLMs and whether procedural deterrents mitigate these effects. Mean test scores were modestly higher following ChatGPT’s release, while cheating-related behaviors increased substantially. Integrity warnings reduced cheating-related behaviors without decreasing scores, whereas facial validation reduced both behaviors and performance. Together, these findings suggest that LLM-related validity threats manifest more strongly in behavior than in score inflation and that targeted deterrents can effectively reduce problematic behaviors.
科研通智能强力驱动
Strongly Powered by AbleSci AI