CASEBOOK

Browse / Model Evasion / AIID-466

AI-Generated Text Detectors Reportedly Misclassified Human and AI-Written Text

AIID-466 · 2023-01-31 real-world incident

Class
Model Evasion (filed by a model that read the source)
Technique
Real-world incidents
Affected
Openai, Edward Tian (incident)

What happened, in short

Publicly tested AI-text detectors, including OpenAI's classifier, produced false results and were easily evaded by modified text. (Our summary, written by a model. The source text is below.)

Details

AI-generated text detectors, including OpenAI’s AI Text Classifier, GPTZero, and Originality.ai, were publicly tested in early 2023 and reportedly produced false positives, false negatives, and easy evasion. OpenAI said its classifier identified only 26% of AI-written samples as likely AI-written and falsely labeled 9% of human-written text.

References

Source: AI Incident Database, used under CC BY-SA 4.0.