AI Detectors Crumble Against Fine-Tuned Models: Pangram Falls to 3% Accuracy, GPTZero to 0%
Dr_WhoDo · reddit · 2026-09-16
A Reddit post exposes a major blind spot in AI detection tools.
- Pangram and GPTZero claim 97% and 91% accuracy against style-prompted LLMs with near-zero false positives
- But against models fine-tuned to mimic specific authors or bodies of work, Pangram's accuracy collapses to 3% and GPTZero's to 0%
- Commercially available models can already be fine-tuned to any author's style, and detectors have no answer
The author also cites the Stony Brook / Columbia Law / MIT study Readers Prefer Outputs of AI Trained on Copyrighted Books over Expert Human Writers — readers actually prefer the fine-tuned AI prose. Conclusion: detectors' headline numbers only hold against average LLM output, and vendors are masking a fundamental flaw in their solutions.
More from Models
- Open-source VoxLock tops Krisp voice isolation benchmark with 4.95% WER — Arbaz Siddiqui · 2026-11-02
- Anthropic's 85% inference margin sparks debate on open source's price-pressuring role — code_star · 2026-11-02
- User reports xAI appears to have quietly raised weekly usage limits, dropping his meter from 93% to 47% — XFreeze · 2026-11-02
- Meituan's LongCat-2.5 lands as 1.6T MoE agent model, with DeepSeek copycat speculation — teortaxesTex · 2026-11-02
- Asked what AI firms want you to stop owning, Claude said: your own compute — alex_verem · 2026-11-02
- Is Hermes Still Worth It? Codex's $200 Tier Now Overlaps Most of Its Value — Foreign_Lead_3582 · 2026-11-02