NTU to stop using 'fundamentally unreliable' AI detector from 2027

NTU is ditching its institutional AI detector. The university told faculty on Aug 13 that it will deactivate the tool by the end of this semester — Dec 31 — and stop using it entirely from 2027, describing automated detection as "fundamentally unreliable" and lacking "empirical validity."

The move follows a turbulent 2025, when three NTU students were accused of misusing generative AI for false and inaccurate citations, and one student's zero mark was upheld on appeal. Dr Tan Seng Chee, associate vice-provost for education transformation, said such tools cannot distinguish genuine misconduct from the responsible use of AI that the curriculum increasingly demands.

From here, NTU will no longer treat AI probability scores as evidence of academic misconduct — a stance it says aligns with emerging best practice at local and international universities. The shift is toward assessment designs that evaluate student learning through transparent, process-oriented documentation, with faculty instructed to treat detector flags as non-evidence and lean on standardised AI disclosure forms instead.

The decision also clears a contradiction: NTU now gives all faculty and students access to its own AI enterprise suite, and running automated surveillance alongside it sent mixed signals about how AI should actually be used. Dr Tan framed the change as a move from passive surveillance toward responsible AI integration and a culture of mutual trust.

Why it matters for Singapore: As one of the country's premier research universities, NTU's stance is a bellwether for academic AI policy nationwide — CNA has already asked other autonomous universities whether they will follow. It also signals a maturing approach in Singapore's AI strategy: teaching responsible use and designing assessment around it, rather than policing students with unreliable detection scores.