Detecting AI-made content becomes a product category

Pangram released Pangram Image 1.0, extending its AI text detector to images. Commenters said its text detection has been good enough to be useful in practice, and several were more willing to believe in detection for images and video than for short text. The top skeptic question was simply 'what's the use case?' Older but related threads show the same unease: an Ask HN proposing a flag for AI-generated articles, where people argue that detectors aren't reliable and the guidelines already ban generated comments, and Cliff Stoll's post about rumors of his death that fed on AI slop.

The pattern: the Cuban story about deepfake scam ads at Facebook is the demand side. When ads, posts and obituaries can be fabricated for pennies, someone gets paid to say what's real. Detection, provenance and moderation tooling are all aimed at the same gap.

The counterpoint is the false-positive problem. A tool that wrongly labels a human's work as generated causes real harm, and that is the main reason people hold back.