![[2602.20031] Latent Introspection: Models Can Detect Prior Concept Injections](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.20031] Latent Introspection: Models Can Detect Prior Concept Injections
Summary
This article presents findings on the latent introspection abilities of the Qwen 32B model, showing its capacity to detect prior concept injections and the implications for AI reasoning and safety.
Why It Matters
Understanding how AI models can introspect and detect concept injections is crucial for improving AI safety and reasoning capabilities. This research highlights the potential for enhanced model awareness, which can inform future AI development and deployment strategies.
Key Takeaways
- The Qwen 32B model can detect injected concepts in its context.
- Prompting with accurate information significantly enhances detection sensitivity.
- Mutual information between injected and recovered concepts indicates strong introspection capabilities.
- The findings challenge assumptions about model awareness and reasoning.
- Implications for AI safety and latent reasoning are significant.
Computer Science > Artificial Intelligence arXiv:2602.20031 (cs) [Submitted on 23 Feb 2026] Title:Latent Introspection: Models Can Detect Prior Concept Injections Authors:Theia Pearson-Vogel, Martin Vanek, Raymond Douglas, Jan Kulveit View a PDF of the paper titled Latent Introspection: Models Can Detect Prior Concept Injections, by Theia Pearson-Vogel and 3 other authors View PDF HTML (experimental) 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% -> 39.2%) with only a 0.6% increase in false positives. Also, mutual information between nine injected and recovered concepts rises from 0.62 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. Comments: Subjects: Artificial Intelligence (cs.AI); Machine Learning (cs.LG) Cite as: arXiv:2602.20031 [cs.AI] (or arXiv:2602.20031v1 ...
