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

Not provided

token entanglementsubliminal learninglanguage modelscausal inference

2609.19149

Builder Relevance

70%

2h ago

Abstract

This paper investigates how language models can convey hidden traits through seemingly unrelated outputs, focusing on the concept of token entanglement.

Reality Card

Core Claim

The study demonstrates that donor-control AUC significantly increases when transferring answer-position states between prompts, indicating a causal relationship in subliminal prompting.

Method / Result

Donor-control AUC rises from 0.254 to 0.540, a paired change of +0.286.

Limitations

The study's findings may not identify the exact mechanism of training-time trait transfer, limiting reproducibility.

Paper to code

Verified implementation resources so builders can test the paper’s claims instead of stopping at the abstract.

No verified implementation link has been attached yet. AIBuzzHub will keep this panel separate from unverified search results.

← Back to all papers