Pangram says its new AI text detector makes only one mistake per 24,000 documents

Editor's take

Pangram claims its latest AI text detector, Pangram 4, achieves an unprecedented 99.66% accuracy in identifying AI-generated content, with a reported false positive rate of one per 24,000 documents. This development is significant for academic institutions and content platforms grappling with the proliferation of AI-authored submissions and articles, offering a potential solution to maintain academic integrity and combat misinformation. The enhanced resistance to "humanizer" tools, which have previously undermined detection efforts, further strengthens its utility.

The substantial price increase for API access, ranging from two to tenfold, suggests Pangram anticipates strong demand from entities prioritizing robust AI detection. This move could create a tiered market for AI detection, with higher costs potentially limiting adoption for smaller organizations or individual researchers. Future developments will likely focus on the long-term efficacy of Pangram 4 against increasingly sophisticated AI writing models and the competitive landscape as other companies respond to this claimed leap in detection capability.