# Multimodal Policy Internalization for Conversational Agents

arXiv: https://arxiv.org/abs/2510.09474 · HTML: https://mlsanity.com/paper/2510.09474/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Placed in General LLM agent frameworks and benchmarks, one of 16 directions in LLM agents and tool use.
- Builds on 35 of its 39 references that are in the map.
- First posted on arXiv 10 Oct 2025.

## Placed in

- LLM agents and tool use / [General LLM agent frameworks and benchmarks](https://mlsanity.com/agents/general-llm-agent-frameworks-and-benchmarks/)

## Builds on (most cited first, 30 of 35)

- [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://mlsanity.com/paper/2402.03300/) (2024)
- [Qwen2.5-VL Technical Report](https://mlsanity.com/paper/2502.13923/) (2025)
- [BERTScore: Evaluating Text Generation with BERT](https://mlsanity.com/paper/1904.09675/) (2019)
- [Prefix-Tuning: Optimizing Continuous Prompts for Generation](https://mlsanity.com/paper/2101.00190/) (2021)
- [DreamBooth: Fine Tuning Text-to-Image Diffusion Models for Subject-Driven Generation](https://mlsanity.com/paper/2208.12242/) (2022)
- [DAPO: An Open-Source LLM Reinforcement Learning System at Scale](https://mlsanity.com/paper/2503.14476/) (2025)
- [An Image is Worth One Word: Personalizing Text-to-Image Generation using Textual Inversion](https://mlsanity.com/paper/2208.01618/) (2022)
- [MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmark](https://mlsanity.com/paper/2406.01574/) (2024)
- [CLEVR: A Diagnostic Dataset for Compositional Language and Elementary Visual Reasoning](https://mlsanity.com/paper/1612.06890/) (2016)
- [τ-bench: A Benchmark for Tool-Agent-User Interaction in Real-World Domains](https://mlsanity.com/paper/2406.12045/) (2024)
- [MMMU-Pro: A More Robust Multi-discipline Multimodal Understanding Benchmark](https://mlsanity.com/paper/2409.02813/) (2024)
- [WildGuard: Open One-Stop Moderation Tools for Safety Risks, Jailbreaks, and Refusals of LLMs](https://mlsanity.com/paper/2406.18495/) (2024)
- [Learning to Compress Prompts with Gist Tokens](https://mlsanity.com/paper/2304.08467/) (2023)
- [Deliberative Alignment: Reasoning Enables Safer Language Models](https://mlsanity.com/paper/2412.16339/) (2024)
- [LLMLingua: Compressing Prompts for Accelerated Inference of Large Language Models](https://mlsanity.com/paper/2310.05736/) (2023)
- [In-context Autoencoder for Context Compression in a Large Language Model](https://mlsanity.com/paper/2307.06945/) (2023)
- [Compressing Context to Enhance Inference Efficiency of Large Language Models](https://mlsanity.com/paper/2310.06201/) (2023)
- [Rephrasing the Web: A Recipe for Compute and Data-Efficient Language Modeling](https://mlsanity.com/paper/2401.16380/) (2024)
- [Prompt Compression and Contrastive Conditioning for Controllability and Toxicity Reduction in Language Models](https://mlsanity.com/paper/2210.03162/) (2022)
- [NoisyRollout: Reinforcing Visual Reasoning with Data Augmentation](https://mlsanity.com/paper/2504.13055/) (2025)
- [GTA: A Benchmark for General Tool Agents](https://mlsanity.com/paper/2407.08713/) (2024)
- [Prompt Compression for Large Language Models: A Survey](https://mlsanity.com/paper/2410.12388/) (2024)
- [Yo'LLaVA: Your Personalized Language and Vision Assistant](https://mlsanity.com/paper/2406.09400/) (2024)
- [MIA-Bench: Towards Better Instruction Following Evaluation of Multimodal LLMs](https://mlsanity.com/paper/2407.01509/) (2024)
- [MyVLM: Personalizing VLMs for User-Specific Queries](https://mlsanity.com/paper/2403.14599/) (2024)
- [Learning to Compress Prompt in Natural Language Formats](https://mlsanity.com/paper/2402.18700/) (2024)
- [R1-ShareVL: Incentivizing Reasoning Capability of Multimodal Large Language Models via Share-GRPO](https://mlsanity.com/paper/2505.16673/) (2025)
- [Injecting Domain-Specific Knowledge into Large Language Models: A Comprehensive Survey](https://mlsanity.com/paper/2502.10708/) (2025)
- [Recent Advances of Multimodal Continual Learning: A Comprehensive Survey](https://mlsanity.com/paper/2410.05352/) (2024)
- [YoChameleon: Personalized Vision and Language Generation](https://mlsanity.com/paper/2504.20998/) (2025)

## Built on by (most cited first, 2 of 2)

- [Analyzing and Internalizing Complex Policy Documents for LLM Agents](https://mlsanity.com/paper/2510.11588/) (2025)
- [Brick-Composer: Using MLLMs for Assembly with Diverse Bricks](https://mlsanity.com/paper/2606.05445/) (2026)

## Abstract (arXiv)

Modern conversational agents like ChatGPT and Alexa+ rely on predefined policies specifying metadata, response styles, and tool-usage rules. As these LLM-based systems expand to support diverse business and user queries, such policies, often implemented as in-context prompts, are becoming increasingly complex and lengthy, making faithful adherence difficult and imposing large fixed computational costs. With the rise of multimodal agents, policies that govern visual and multimodal behaviors are critical but remain understudied. Prior prompt-compression work mainly shortens task templates and demonstrations, while existing policy-alignment studies focus only on text-based safety rules. We introduce Multimodal Policy Internalization (MPI), a new task that internalizes reasoning-intensive multimodal policies into model parameters, enabling stronger policy-following without including the policy during inference. MPI poses unique data and algorithmic challenges. We build two datasets spanning synthetic and real-world decision-making and tool-using tasks and propose TriMPI, a three-stage training framework. TriMPI first injects policy knowledge via continual pretraining, then performs supervised finetuning, and finally applies PolicyRollout, a GRPO-style reinforcement learning extension that augments rollouts with policy-aware responses for grounded exploration. TriMPI achieves notable gains in end-to-end accuracy, generalization, and robustness to forgetting. As the first work on multimodal policy internalization, we provide datasets, training recipes, and comprehensive evaluations to foster future research. Project page: https://mikewangwzhl.github.io/TriMPI.