America Has a Pangram Problem

Pangram, an AI-detection program, has emerged as the go-to tool for identifying AI-generated writing, sparking controversies in publishing, journalism, and academia. In March, a horror novel from a major publisher was pulled after Pangram flagged portions as AI-written, while the tool has also raised concerns about AI-assisted articles in *The New York Times*, award-winning short stories, and even Pope Leo XIV’s encyclical. Universities and scientific associations now rely on Pangram to vet student work and research papers, amplifying its role amid rising fears of AI-generated content. The tool’s accuracy is widely praised, with Pangram claiming a false-positive rate of just one in 10,000 texts, a claim supported by independent studies like one from the University of Chicago. However, its ability to detect AI text remains imperfect, with false-negative rates estimated as high as one in 70, meaning some AI-generated content slips through as human-written. CEO Max Spero acknowledges the responsibility in making such claims, insisting Pangram only flags AI text when ‘extremely confident.’ The arms race between AI detectors like Pangram and AI developers—who refine models like ChatGPT and Claude to sound more human—complicates detection efforts. Meanwhile, ‘humanizer’ tools like Walter Writes AI further obscure AI-generated text by tweaking phrasing and grammar. Testing showed Pangram failed to detect AI text processed through such tools, raising concerns about false reassurance in its assessments. Critics argue Pangram’s growing influence risks fueling witch hunts, as its judgments can derail careers and reputations without perfect accuracy. While the tool has improved detection capabilities, its limitations—particularly in false negatives—pose ethical and practical challenges in an era where AI-generated content is increasingly indistinguishable from human work.