Building Auditable AI Text Detectors with Sparse Autoencoders
We present a pipeline that converts a fine-tuned neural detector into a classifier that operates on human-readable concepts.
ProofPaper ↗
Key points
- 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 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 find that our SAEs recover latent concepts that correspond to highly specific and predictive linguistic "tells" of AI-generated text.
Sources (1)
- [1]Building Auditable AI Text Detectors with Sparse AutoencodersarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 21, 04:41 PM
We present a pipeline that converts a fine-tuned neural detector into a classifier that operates on human-readable concepts.
AI-generated text detectors are increasingly deployed in consequential settings---peer review, education, admissions, journalism, and content moderation---where their outputs directly affect individuals.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 20, 2026Tropical Cyclone Attributes in Sparse Autoencoder Features of an AI Weather Model
- Oct 20, 2026Where Do Legal Factors Live in LLMs?
- Oct 20, 2026Decision Sufficiency of Learned Representation Interfaces for Safe Generative Control
- Oct 20, 2026The Router Already Knows
- Oct 20, 2026Not All Features Are Created Equal: A Mechanistic Study of Vision-Language-Action Models
- Oct 19, 2026Shaped SAEs: Expanding the Toolkit for Mechanistic Interpretability