WorkJuly 19, 2026via The Decoder
AI text detectors struggle when language models mimic an author's style
Why it matters
AI text detectors—sold as safeguards against academic fraud and plagiarism—have significant blind spots when LLMs mimic author style. This matters for institutions relying on these tools for compliance and integrity, and for the credibility of detection as a defensive strategy.
Key signals
- Epoch AI tested Pangram, GPTZero, and Originality.ai
- 18% of AI-generated passages went undetected overall
- 48% miss rate for scientific writing genre
- Style imitation significantly degrades detector accuracy
The hook
48% miss rate on scientific papers. AI detectors are failing at their core job—and nobody's talking about the audit gap.
Epoch AI tested three leading AI text detectors (Pangram, GPTZero, and Originality.ai) using style-imitated texts. Up to 18 percent of AI-generated passages went undetected. For scientific writing, the miss rate climbed as high as 48 percent, the very genre where these detectors likely see the most real-world use.