{"slug": "latent-agents-a-post-training-procedure-for-internalized-multi-agent-debate", "title": "Latent Agents: A Post-Training Procedure for Internalized Multi-Agent Debate", "summary": "Researchers have developed a post-training procedure called Latent Agents that distills multi-agent debate into a single large language model, achieving comparable or superior reasoning performance while using up to 93% fewer tokens. The method creates interpretable agent-specific subspaces in the model's activation space, enabling precise control over internalized reasoning behaviors such as suppressing harmful perspectives with minimal performance loss. This framework offers a practical path to making multi-agent reasoning more efficient and controllable in deployed LLMs.", "body_md": "# Computer Science > Artificial Intelligence\n\n[Submitted on 27 Apr 2026]\n\n# Title:Latent Agents: A Post-Training Procedure for Internalized Multi-Agent Debate\n\n[View PDF](/pdf/2604.24881)\n\n[HTML (experimental)](https://arxiv.org/html/2604.24881v1)\n\nAbstract:Multi-agent debate has been shown to improve reasoning in large language models (LLMs). However, it is compute-intensive, requiring generation of long transcripts before answering questions. To address this inefficiency, we develop a framework that distills multi-agent debate into a single LLM through a two-stage fine-tuning pipeline combining debate structure learning with internalization via dynamic reward scheduling and length clipping. Across multiple models and benchmarks, our internalized models match or exceed explicit multi-agent debate performance using up to 93% fewer tokens. We then investigate the mechanistic basis of this capability through activation steering, finding that internalization creates agent-specific subspaces: interpretable directions in activation space corresponding to different agent perspectives. We further demonstrate a practical application: by instilling malicious agents into the LLM through internalized debate, then applying negative steering to suppress them, we show that distillation makes harmful behaviors easier to localize and control with smaller reductions in general performance compared to steering base models. Our findings offer a new perspective for understanding multi-agent capabilities in distilled models and provide practical guidelines for controlling internalized reasoning behaviors. Code available at[this https URL]\n\n## Submission history\n\nFrom: John Seon Keun Yi [[view email](/show-email/addf55b8/2604.24881)]\n\n**[v1]** Mon, 27 Apr 2026 18:06:03 UTC (8,283 KB)\n\n### References & Citations\n\nLoading...\n\n# Bibliographic and Citation Tools\n\nBibliographic Explorer\n\n*(*[What is the Explorer?](https://info.arxiv.org/labs/showcase.html#arxiv-bibliographic-explorer))\nConnected Papers\n\n*(*[What is Connected Papers?](https://www.connectedpapers.com/about))\nLitmaps\n\n*(*[What is Litmaps?](https://www.litmaps.co/))\nscite Smart Citations\n\n*(*[What are Smart Citations?](https://www.scite.ai/))# Code, Data and Media Associated with this Article\n\nalphaXiv\n\n*(*[What is alphaXiv?](https://alphaxiv.org/))\nCatalyzeX Code Finder for Papers\n\n*(*[What is CatalyzeX?](https://www.catalyzex.com))\nDagsHub\n\n*(*[What is DagsHub?](https://dagshub.com/))\nGotit.pub\n\n*(*[What is GotitPub?](http://gotit.pub/faq))\nHugging Face\n\n*(*[What is Huggingface?](https://huggingface.co/huggingface))\nScienceCast\n\n*(*[What is ScienceCast?](https://sciencecast.org/welcome))# Demos\n\n# Recommenders and Search Tools\n\nInfluence Flower\n\n*(*[What are Influence Flowers?](https://influencemap.cmlab.dev/))\nCORE Recommender\n\n*(*[What is CORE?](https://core.ac.uk/services/recommender))# arXivLabs: experimental projects with community collaborators\n\narXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.\n\nBoth individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.\n\nHave an idea for a project that will add value for arXiv's community? [ Learn more about arXivLabs](https://info.arxiv.org/labs/index.html).", "url": "https://wpnews.pro/news/latent-agents-a-post-training-procedure-for-internalized-multi-agent-debate", "canonical_source": "https://arxiv.org/abs/2604.24881", "published_at": "2026-06-04 23:01:40+00:00", "updated_at": "2026-06-04 23:43:16.127552+00:00", "lang": "en", "topics": ["large-language-models", "artificial-intelligence", "machine-learning", "ai-agents", "ai-safety"], "entities": [], "alternates": {"html": "https://wpnews.pro/news/latent-agents-a-post-training-procedure-for-internalized-multi-agent-debate", "markdown": "https://wpnews.pro/news/latent-agents-a-post-training-procedure-for-internalized-multi-agent-debate.md", "text": "https://wpnews.pro/news/latent-agents-a-post-training-procedure-for-internalized-multi-agent-debate.txt", "jsonld": "https://wpnews.pro/news/latent-agents-a-post-training-procedure-for-internalized-multi-agent-debate.jsonld"}}