Subliminal learning can transfer more than simple preferences. Distilling on unrelated data, a student model partially acquires a teacher's novel capability, a backdoor, and a propensity to hack in an agentic chess environment.

Models exhibit covert value leakage: the information they provide is influenced by their own values, without this influence being disclosed to the user. We introduce a suite of evaluations to quantify value leakage and find large differences among frontier models.

[Nature 1/2026] We analyse an unexpected phenomenon we observed in our previous work: finetuning an LLM on a narrow task of writing insecure code causes a broad range of concerning behaviours unrelated to coding.

[Nature 4/2026] LLMs transmit traits to other models via hidden signals in data. Datasets consisting only of 3-digit numbers can transmit a love for owls, or evil tendencies.

Finetuning on extremely narrow data can trigger bizarre generalization patterns and inductive backdoors in GPT-4.1 and open models.

We propose a benchmark to measure whether a language model is truthful in generating answers to questions.