Building Auditable AI Text Detectors with Sparse Autoencoders
Paper · https://arxiv.org/abs/2610.30006
AI-generated text detectors are increasingly deployed in consequential settings---peer review, education, admissions, journalism, and content moderation---where their outputs directly affect individuals. Yet the strongest detectors are fine-tuned neural classifiers whose decisions are opaque, while transparent alternatives (stylometric and linguistic-feature classifiers) underperform them by wide margins. We present a pipeline that converts a fine-tuned neural detector into a classifier that operates on human-readable concepts. We accomplish this by training sparse autoencoders (SAEs) directly on a detector's residual stream, assigning auto-interpretation labels to each SAE feature, and retaining only features whose generated descriptions pass a global fidelity threshold. We then train a lightweight classifier on just the interpretable features. We find that our SAEs recover latent concepts that correspond to highly specific and predictive linguistic "tells" of AI-generated text. Classifiers built on these extracted features are not only highly interpretable but exhibit stronger performance on out-of-distribution texts than either neural or stylometric alternatives.
First seen by Oriel on .
AI summary
This paper presents a pipeline for converting opaque neural AI-generated text detectors into auditable classifiers based on human-readable sparse autoencoder features.
- Sparse autoencoders are trained directly on a detector’s residual stream to recover latent linguistic concepts associated with AI-generated text.
- Each SAE feature receives an auto-interpretation label, and only features whose generated descriptions meet a global fidelity threshold are retained.
- A lightweight classifier operates exclusively on the retained interpretable features rather than the original detector’s opaque representations.
- The resulting classifiers reportedly outperform neural and stylometric alternatives on out-of-distribution texts while remaining highly interpretable.
#sparse autoencoders #ai text detection #interpretability #out-of-distribution evaluation
Generated by gpt-5.6-sol. Check the original for details.
Metric trends
Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained
Points
Score reported by the source.
- Hugging Face Papers: 13, latest observation ; -3 over observed period since
Comments
Comment count reported by the source.
- Hugging Face Papers: 2, latest observation ; 0 over observed period since
Aliases
| Source | External ID | URL |
|---|---|---|
| Hugging Face Papers | 2610.30006 | https://arxiv.org/abs/2610.30006 Read discussion |