# Out-of-Distribution Generalization of Risk Aversion in Language Models

arXiv: https://arxiv.org/abs/2607.02755 · HTML: https://mlsanity.com/paper/2607.02755/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 21 of its 38 references that are in the map.
- First posted on arXiv 2 Jul 2026.

## Builds on (most cited first, 21 of 21)

- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [LoRA: Low-Rank Adaptation of Large Language Models](https://mlsanity.com/paper/2106.09685/) (2021)
- [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [Fine-Tuning Language Models from Human Preferences](https://mlsanity.com/paper/1909.08593/) (2019)
- [Representation Engineering: A Top-Down Approach to AI Transparency](https://mlsanity.com/paper/2310.01405/) (2023)
- [Steering Llama 2 via Contrastive Activation Addition](https://mlsanity.com/paper/2312.06681/) (2023)
- [Steering Language Models With Activation Engineering](https://mlsanity.com/paper/2308.10248/) (2023)
- [SFT Memorizes, RL Generalizes: A Comparative Study of Foundation Model Post-training](https://mlsanity.com/paper/2501.17161/) (2025)
- [Exploring Length Generalization in Large Language Models](https://mlsanity.com/paper/2207.04901/) (2022)
- [Are We Done with MMLU?](https://mlsanity.com/paper/2406.04127/) (2024)
- [Natural Emergent Misalignment from Reward Hacking in Production RL](https://mlsanity.com/paper/2511.18397/) (2025)
- [Tell me about yourself: LLMs are aware of their learned behaviors](https://mlsanity.com/paper/2501.11120/) (2025)
- [GLUE-X: Evaluating Natural Language Understanding Models from an Out-of-distribution Generalization Perspective](https://mlsanity.com/paper/2211.08073/) (2022)
- [Utility Engineering: Analyzing and Controlling Emergent Value Systems in AIs](https://mlsanity.com/paper/2502.08640/) (2025)
- [LLM economicus? Mapping the Behavioral Biases of LLMs via Utility Theory](https://mlsanity.com/paper/2408.02784/) (2024)
- [Decision-Making Behavior Evaluation Framework for LLMs under Uncertain Context](https://mlsanity.com/paper/2406.05972/) (2024)
- [Steering Out-of-Distribution Generalization with Concept Ablation Fine-Tuning](https://mlsanity.com/paper/2507.16795/) (2025)
- [Evaluating and Aligning Human Economic Risk Preferences in LLMs](https://mlsanity.com/paper/2503.06646/) (2025)
- [Spurious Correlation Learning in Preference Optimization: Mechanisms, Consequences, and Mitigation via Tie Training](https://mlsanity.com/paper/2605.11134/) (2026)
- [Risk Profiling and Modulation for LLMs](https://mlsanity.com/paper/2509.23058/) (2025)

## Built on by (most cited first, 1 of 1)

- [Character Training for Risk-Averse Agents](https://mlsanity.com/paper/2609.38093/) (2026)

## Abstract (arXiv)

Training AIs to be risk-averse in resources could offer a failsafe in the event that AIs turn out misaligned. Misaligned but risk-averse AIs would tend to prefer low-risk, low-reward strategies like cooperation over high-risk, high-reward strategies like rebellion, limiting the downsides of any misalignment. But we can only feasibly train AIs to be risk-averse on low-stakes gambles, and we will only be safe if their risk aversion generalizes to astronomically-high-stakes gambles. Will it? To shed light on this question, we introduce RiskAverseOOD: a benchmark for measuring how well risk aversion generalizes out of distribution. We then offer some initial results. Using a variety of methods to make Qwen3-8B choose risk-aversely when the stakes are low, we find that we can induce substantial risk aversion when the stakes are astronomically high. From a baseline 2% rate of choosing a safe `Cooperate' option, we see rates around 70% (SFT and tie training) and 52% (DPO). Activation steering scores 78% but hurts capabilities and makes the model excessively risk-averse. We observe similar effects at different scales (Qwen3-1.7B and Qwen3-14B) and across model families (Gemma-3-12B-IT and Llama-3.1-8B-Instruct). Overall, we find that risk aversion learned at low stakes can generalize OOD to astronomically high stakes, though not yet consistently enough to serve as a reliable failsafe. Achieving that level of consistency is an open problem.