AI-detection tools have made huge leaps forward — how good are they?
Nature, Published online: 25 August 2026; doi:10.1038/d41586-026-02569-3 Scientists and publishers are testing a new breed of software that promises impressive accuracy at spotting AI-written text.
AI-detection tools have made significant advancements, with various firms claiming to reliably distinguish between AI-written and human-written text. Pangram Labs, a New York City-based startup, asserts its software can achieve 99.98% accuracy. According to Pangram, around one in eight biomedical articles contained AI-generated text in the previous year.
In June, NeurIPS, a major computer-science conference, rejected 18% of submissions after screening them with Pangram. Pangram's co-founder, Max Spero, aims to help identify AI-written text, stating that failure to address the issue could lead to more people relying on AI. Pangram has been integrated into Substack, a popular blogging platform, enabling readers to detect AI-written posts.
Universities also rely on AI-detection software to combat academic dishonesty. While these tools are generally effective at identifying purely human-written content and help screen out low-quality content, they are not infallible and may struggle with AI-edited writing that blends both human and AI-generated text.
Written by urgent.news from Nature's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
