Pangram's biggest flaw is users turning its scores into public shaming
decoder ·
AI text detection firm Pangram hired journalist Rod Breslau to publicly shame social media users for suspected AI use, though the company later severed ties with him and shifted direction. CEO Max Spero has nevertheless kept using Pangram scores to call out individuals he believes hid their AI involvement. A key weakness is that the service can only indicate whether AI was likely involved, not how it was used—such as whether text was fully generated, edited, or translated. Public shaming implies the writer did not think or work independently, a conclusion the scores cannot support. The article notes its own English translation received a 28 percent AI score, despite every section being produced and edited with the same level of human intention. It also warns that academic institutions reject papers based on such scores, penalizing researchers, especially those writing in a second language, who use AI to improve clarity. A historical parallel is drawn to Alexandre Dumas, who relied on assistants for rough drafts but refined the work himself, showing that detection misses the essence of human creative effort.
decoder