# Subliminal Prompting Beyond Static Geometry: Causal Depth and Multi-Token Confounds

arXiv: https://arxiv.org/abs/2609.19149 · HTML: https://mlsanity.com/paper/2609.19149/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 12 of its 18 references that are in the map.
- First posted on arXiv 20 Jul 2026.

## Builds on (most cited first, 12 of 12)

- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [Locating and Editing Factual Associations in GPT](https://mlsanity.com/paper/2202.05262/) (2022)
- [Transformer Feed-Forward Layers Build Predictions by Promoting Concepts in the Vocabulary Space](https://mlsanity.com/paper/2203.14680/) (2022)
- [Towards Best Practices of Activation Patching in Language Models: Metrics and Methods](https://mlsanity.com/paper/2309.16042/) (2023)
- [Does Circuit Analysis Interpretability Scale? Evidence from Multiple Choice Capabilities in Chinchilla](https://mlsanity.com/paper/2307.09458/) (2023)
- [Subliminal Learning: Language models transmit behavioral traits via hidden signals in data](https://mlsanity.com/paper/2507.14805/) (2025)
- [Towards Understanding Subliminal Learning: When and How Hidden Biases Transfer](https://mlsanity.com/paper/2509.23886/) (2025)
- [Where is the signal in tokenization space?](https://mlsanity.com/paper/2408.08541/) (2024)
- [Subliminal Steering: Stronger Encoding of Hidden Signals](https://mlsanity.com/paper/2604.25783/) (2026)
- [Learning Through Noise: Why Subliminal Learning Works and When It Fails](https://mlsanity.com/paper/2605.23645/) (2026)
- [Channel Location Constrains the Auditability of Subliminal Learning](https://mlsanity.com/paper/2606.22019/) (2026)
- [Covert Trait Propagation Is Representation Alignment: Mechanistic Evidence from Hidden-Channel Distillation](https://mlsanity.com/paper/2607.04432/) (2026)

## Abstract (arXiv)

Subliminal learning shows that language models can transmit a hidden trait through outputs that appear unrelated to it. One proposed explanation, token entanglement, links animal and number tokens through the model's output vocabulary. Yet existing measurements answer different questions: whether outputs co-vary, fixed output vectors align, an answer can be read from a hidden state, or that state causally controls the answer. We measure each separately in a fixed animal-number prompting protocol. From Llama-3.1-8B to 70B, fixed output-vector similarity predicts behavior less well: the paired mean correlation change is -0.080 (95% CI [-0.127, -0.035]). A fixed output-head readout shows no resolved change in normalized depth AUC. To test control, we copy the temporary answer-position state from one number prompt into another at five depths and measure which prompt the final animal score follows. Donor-control AUC rises from 0.254 to 0.540, a paired change of +0.286 (95% CI [+0.272, +0.300]), with increases for all 18 concepts. The contrast remains with exactly eight transformer blocks remaining, while specificity and identity controls remain small or exact. In two Qwen models, scoring every digit in sequence does not recover the positive one-token association. Per-token averaging instead creates a positive pooled association that disappears after controlling number width, revealing a length confound. Thus, fixed geometry, observational readability, causal timing, and multi-token measurement are distinct properties of this frozen prompting channel. They constrain token-level explanations but do not identify the mechanism of training-time trait transfer.