Inverse Language Modeling towards Robust and Grounded LLMs

2025-10-02

AI Summary

Current defenses for large language models (LLMs) are fragmented, lack robustness, and fail to identify harmful inputs via output-driven reverse inference. To address this, we propose Inverse Language Modeling (ILM), the first framework to leverage inverse modeling for enhancing LLM adversarial robustness and intrinsic interpretability. ILM jointly optimizes input perturbation defense and output-based backward reasoning, enabling native detection of toxic triggers, content-grounded explanations, and behavior analyzability—without external classifiers or model fine-tuning. It supports red-teaming evaluations and real-time safety responses. Experiments demonstrate that ILM significantly improves model stability under adversarial attacks while increasing detection accuracy and traceability of latent unsafe inputs. By unifying defensive and explanatory capabilities within a single forward-inverse modeling paradigm, ILM establishes a novel foundation for developing controllable, trustworthy next-generation LLMs.

Technology Category

• Machine Learning: Adversarial Learning & Robustness
• Natural Language Processing: Safety and Robustness
• Computer Vision: Adversarial Attacks & Robustness

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

The current landscape of defensive mechanisms for LLMs is fragmented and underdeveloped, unlike prior work on classifiers. To further promote adversarial robustness in LLMs, we propose Inverse Language Modeling (ILM), a unified framework that simultaneously 1) improves the robustness of LLMs to input perturbations, and, at the same time, 2) enables native grounding by inverting model outputs to identify potentially toxic or unsafe input triggers. ILM transforms LLMs from static generators into analyzable and robust systems, potentially helping RED teaming. ILM can lay the foundation for next-generation LLMs that are not only robust and grounded but also fundamentally more controllable and trustworthy. The code is publicly available at github.com/davegabe/pag-llm.

Problem

• Improving LLM robustness to input perturbations
• Enabling native grounding by inverting model outputs
• Transforming LLMs into analyzable and controllable systems

Innovation

• Inverse Language Modeling framework improves LLM robustness
• Inverts model outputs to identify toxic input triggers
• Transforms LLMs into analyzable and robust systems