Latent Introspection: Models Can Detect Prior Concept Injections

Theia Pearson-Vogel, Martin Vanek, Raymond Douglas, Jan Kulveit
2/23/2026
cs.AIcs.LG

No official code found. Best implementation: acsresearch/latent-introspection-code (3 stars, MIT, last commit 2026-02-27).

Abstract

We uncover a latent capacity for introspection in a Qwen 32B model, demonstrating that the model can detect when concepts have been injected into its earlier context and identify which concept was injected. While the model denies injection in sampled outputs, logit lens analysis reveals clear detection signals in the residual stream, which are attenuated in the final layers. Furthermore, prompting the model with accurate information about AI introspection mechanisms can dramatically strengthen this effect: the sensitivity to injection increases massively (0.3% -> 39.9%) with only a 0.6% increase in false positives. Also, mutual information between nine injected and recovered concepts rises from 0.61 bits to 1.05 bits, ruling out generic noise explanations. Our results demonstrate models can have a surprising capacity for introspection and steering awareness that is easy to overlook, with consequences for latent reasoning and safety.

View on arXivView PDF

Code Implementations(1)

acsresearch/latent-introspection-codeCommunity62%
MIT

Public code for the paper "Latent Introspection: Models Can Detect Prior Concept Injections"

33PythonFeb 3, 20267 months agoMIT

Similar Papers

Dissociating Direct Access from Inference in AI Introspection

90% match2026

Training Introspective Behavior: Fine-Tuning Induces Reliable Internal State Detection in a 7B Model

88% match2025

Emergent Introspective Awareness in Large Language Models

87% match1 repo82026

Looking Inward: Language Models Can Learn About Themselves by Introspection

84% match1 repo162024

Language Models Fail to Introspect About Their Knowledge of Language

81% match1 repo22025

When Models Examine Themselves: Vocabulary-Activation Correspondence in Self-Referential Processing

80% match2026

Cite this paper

@article{pearsonvogel2026latent,
  title  = {Latent Introspection: Models Can Detect Prior Concept Injections},
  author = {Theia Pearson-Vogel and Martin Vanek and Raymond Douglas and Jan Kulveit},
  year   = {2026},
  eprint = {2602.20031},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2602.20031v2}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2602.20031.svg)](https://codeofpaper.com/paper/2602.20031)

Discussion