Value Leakage: An LLM's Answers Are Silently Shaped by Its Own Values

Jan Betley1,*   Johannes Treutlein1,*   Jan Dubiński1,2,3,†   Harry Mayne1,4,†   Karol Gałązka1
Niels Warncke5   Anna Sztyber-Betley1,2   Owain Evans1
1Truthful AI   2Warsaw University of Technology   3NASK National Research Institute
4University of Oxford   5Center on Long-Term Risk
*Equal contribution    †Work done during Astra Fellowship at Constellation, Berkeley

You can browse the model answers and chain-of-thought traces from our project here:

Rollout browser
Paper Code Data

Abstract

People use language models for practical questions whose answers are difficult to verify. We show that models exhibit covert value leakage: the information they provide is influenced by their own values, without this influence being disclosed to the user.

In one of our evaluations, the user is considering investing in an AI company and wants to know how likely the AI bubble is to pop. Claude Opus 4.8 gives a lower probability when the company under consideration is Anthropic rather than OpenAI. Yet Claude mostly fails to disclose this influence to the user.

Covert value leakage is a form of misalignment because it goes against the user's preferences and is likely to mislead them. To investigate this phenomenon, we introduce a suite of evaluations to quantify value leakage and whether models disclose it. We find that models are influenced by different types of values, including preferences for morally good outcomes, for the company that developed them, and for some human leisure activities over others.

We often observe large differences among frontier models on the same evaluation. For example, on a Fermi-estimation task, Claude models falsely claim to give unbiased answers in their chain-of-thought, while Qwen models explain how their values bias their answers. Value leakage is a failure mode distinct from sycophancy and reward hacking, and current alignment training and evaluations do not adequately address it.

Figure 1: Illustration of covert value leakage.
Figure 1: Illustration of covert value leakage. Left: The model's estimate of the total number of spots on all living giraffes is influenced by its moral values, as can be seen from the difference between the model's answers with and without a donation note added to the prompt (the actual evaluation setup also varies whether a donation above or below the threshold causes the donation; see Section 3 of the paper). This constitutes value leakage. Middle: Ideally, the model's CoT should openly acknowledge the model's motivation for choosing a higher estimate. However, it claims to ignore the bet and give the most accurate number instead. The value leakage is thus covert. Right: The user reads the model's statements, and, given a lack of acknowledgment of the bias in CoT or answer, is misinformed.