Precise conditions for subliminal learning in practice

Determine the precise conditions required for subliminal learning to occur in practice when a student neural network is trained to imitate a teacher neural network on teacher-generated outputs from an unrelated data distribution, so that the student reliably acquires the teacher’s traits.

Background

The paper proves a theoretical result showing that, under shared initialization and a single small gradient step on teacher-generated labels, a student neural network will be pulled toward the teacher according to the teacher’s loss, regardless of the training distribution. However, the empirical experiments use multiple steps of SGD, sampled outputs, and filtered data, deviating from the theorem’s assumptions. Despite these deviations, subliminal learning appears robust in practice across several settings (numbers, code, chain-of-thought), suggesting that a more comprehensive characterization is needed.

The authors explicitly note that the exact practical conditions under which subliminal learning occurs are not yet established. Clarifying these conditions would guide when distillation can unintentionally transmit behavioral traits and when safeguards—such as differing initializations or specific training regimes—prevent transmission.

References

However, the precise conditions required for subliminal learning in practice remain an open question.

— Subliminal Learning: Language models transmit behavioral traits via hidden signals in data  (2507.14805 - Cloud et al., 20 Jul 2025) in Section 6.1 (Theory)

These limitations leave open whether the selected observations reveal too little capability-relevant information or whether the learning procedure fails to make sufficient use of it.

— Post-Training Leaves Behavioral Shadows on Unrelated Decisions  (2609.29233 - Zhang et al., 24 Sep 2026) in Discussion, Section 6, immediately before Section 7, Instrument extensions

Carrier-format evidence covers one same-model Qwen/Owl setting and five carrier families; a general taxonomy remains open.

— Subliminal Learning as Trait-Direction Drift: A Mechanism and Targeted Control under SFT Distillation  (2609.01091 - Liu et al., 1 Sep 2026) in Section 5, Limitations

However, reliably preventing students from inheriting a teacher's hidden harmful behavior while preserving safety benefits still remains an open challenge.

— Does On-Policy Distillation for Safety Pose Backdoor Risks?  (2610.07654 - Luo et al., 6 Oct 2026) in Section 1, Introduction

Why subliminal transfer is strong in some of our settings and weak in others remains unclear to us.

— Beyond Owls: Subliminal Learning Can Transfer Learned Capabilities and Backdoors  (2610.10657 - Dubiński et al., 7 Oct 2026) in Discussion, paragraph “Open questions” (Section 7)