greCAPTCHA: Testing Whether Authors Truly Understand Their Own Papers to Detect AI Ghostwriting
Amid a flood of AI-assisted and AI-generated submissions, author bylines no longer reliably signal genuine expertise. Researcher Atoosa Kasirzadeh proposes greCAPTCHA: instead of passively inspecting text, the system dynamically generates multi-level questions about a submitted manuscript to test the author's actual understanding. A prototype has been implemented, with automated scoring reaching AUC 0.90 in predicting true authorship.
Confirmed
- greCAPTCHA, proposed by Atoosa Kasirzadeh, shifts evaluation from the textual product to the process of intellectual production, measuring the construct of "verification capability"—the domain knowledge and reasoning needed to critically assess and defend one's own contribution.
- The system dynamically generates multi-level questions tailored to a submitted manuscript and automatically produces an assessment report from the author's answers.
- Evaluation was based on an empirical user study and semi-structured interviews with 31 active researchers; automated scoring achieved AUC 0.90 in predicting whether a participant was the true author.
Why it matters
- Traditional AI-text detectors are passive and easily circumvented; greCAPTCHA instead asks whether the author truly understands their own paper, offering a new paradigm for gatekeeping in peer review venues, universities, and education.
- Participants in the user study reported positive experiences, feeling the questions captured genuine authorship, while also identifying weaknesses and suggesting improvements before deployment—indicating a promising direction that still needs refinement.
2026-09-17 ~ 2026-09-17 · 7 related posts
Primary sources
- AI-Generated Submissions Are Flooding Peer Review — A Researcher Proposes Evaluating Process, Not Prose — Dr_Atoosa · 2026-09-17
- greCAPTCHA: testing manuscript-specific understanding to catch AI-written papers — Dr_Atoosa · 2026-09-17
- [source] greCAPTCHA Proposes Measuring Whether Authors Actually Understand Their Own Manuscripts — Dr_Atoosa · 2026-09-17
- Implemented greCAPTCHA Dynamically Generates Multi-Level Questions to Verify Manuscript Authorship — Dr_Atoosa · 2026-09-17
- [source] greCAPTCHA verifies manuscript authorship with 0.90 AUC in user study — Dr_Atoosa · 2026-09-17
- Automated authorship scoring hits AUC 0.90 in verifying real manuscript authors — Dr_Atoosa · 2026-09-17
1 near-duplicate retellings: Dr_Atoosa