As artificial-intelligence text floods the internet, a $100 billion industry of verification has sprung up almost overnight — and Pangram, one of its most prominent players, says the entire framing is wrong. CEO Max Spero argues that the question everyone keeps asking, whether a piece of writing is real or fake, is the wrong question, and that the industry's obsession with binary labels is producing tools that mislead the very institutions trying to use them responsibly.
Spero's core point is that AI writing is no longer a discrete category but a spectrum. Authors increasingly draft with models, edit their output by hand, or run human text through machine polish, producing hybrid work that belongs fully to neither bucket. A detector that must answer 'human or machine' collapses that continuum into a coin flip, and the resulting confidence scores — often presented with scientific precision — are far less reliable than buyers assume. Pangram has instead built its technology around estimating how much of a document was AI-generated, treating detection as a matter of degree rather than a verdict.
The stakes reach far beyond academia, where the debate began. Publishers, employers, legal teams, and platforms are all now adjudicating authorship, frequently with little understanding of how the classifiers work or how often they err — and false accusations of AI use have already produced documented cases of punished students and fired workers. Spero contends that honest uncertainty, transparent probability reporting, and human review are the only defensible path forward.
The larger lesson is uncomfortable but familiar: in the arms race between generation and detection, certainty is the first casualty, and the companies that admit it may ultimately be the ones institutions trust.
