Claude can identify which human wrote a text, and de-anonymization is progressing fast
najoungkim · x · 2026-11-20
Najoung Kim quotes a new quantification study with the joke that she can finally have her signature sign her reviews.
The underlying finding: from a single text, Claude can tell whether it was written by AI or a human — with no tools, internet, or candidate list. If AI, it can identify which lab; if human, it can often identify which human. The researchers note this de-anonymization capability appears to be progressing fast and are quantifying it.
More from Models
- Leaker teases: Google's upcoming models have 'cracked the code' — iruletheworldmo · 2026-11-20
- Qwen4 Max lands on PostTrainBench at 36.7%, edging below GLM 5.3 Flash — maksym_andr · 2026-11-20
- PostTrainBench data: US-China model gap of ~8% is nothing unusual — maksym_andr · 2026-11-20
- Cost-vs-quality chart puts Gemini 4 Neon on the CursorBench Pareto frontier — zacharynado · 2026-11-20
- AI model proves optimal solutions to long-open MIPLIB optimization problems — tdhopper · 2026-11-20
- Claude 3 Sonnet spontaneously writes a full Latin hymn at 7:59 PM — repligate · 2026-11-20