Study: AI Disclosure Jumps to 66% in Course Git Repos, But Attribution Lags

Researchers from Cornell Tech analyzed 203 GitHub repositories and 23,065 student commits across three cohorts (Fall 2022, 2023, 2025) of a graduate-level HCI prototyping course. They tracked two levels of AI accountability: disclosure (naming an AI tool) and attribution (crediting a specific artifact or task to AI). Over the three semesters, disclosure grew from 0% to 66% of repositories, as students increasingly named tools like ChatGPT and GitHub Copilot. However, explicit attribution – specifying what the AI actually contributed – remained a minority practice. The widening gap between disclosure and attribution reveals where accountability is missing even among students who are transparent about using AI.

The authors argue that traditional disclosure norms are built for episodic, identifiable AI use, but the 'vibe-coding' era is characterized by continuous, ambient co-creation where AI is embedded in templates and workflows. By 2025, AI had become infrastructure in the course, making it harder to isolate individual contributions. The study calls for new frameworks that go beyond simple disclosure to capture genuine co-thinking accountability – for example, requiring students to document prompts, iterations, and decisions alongside code. As AI tools become ubiquitous in professional development, this accountability gap has direct implications for software teams, hiring practices, and academic integrity policies.