Frontier Models Identify AI and Human Authors via Zero-Shot Prompting
November 20, 2026
Frontier models like Claude Fable 5.2 and GPT-6.1 Sol can identify if a text is human or AI-written using only a single text and a zero-shot prompt. Fable 5.2 correctly identified 39% of public peer reviews from late 2026 and 69% of active Redditor usernames.
HOW THIS AFFECTS YOU
●
builderYou should assume your model's outputs can be traced back to specific authors or labs through reasoning.
●
policyThis capability increases the difficulty of maintaining anonymity in digital communications.