Preemptive Detection and Steering of LLM Misalignment via Latent Reachability
Fuente:
arXiv
Guardado en:
| Autores principales: | , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2025
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866911176528494592 |
|---|---|
| author | Karnik, Sathwik Bansal, Somil |
| author_facet | Karnik, Sathwik Bansal, Somil |
| contents | Large language models (LLMs) are now ubiquitous in everyday tools, raising urgent safety concerns about their tendency to generate harmful content. The dominant safety approach -- reinforcement learning from human feedback (RLHF) -- effectively shapes model behavior during training but offers no safeguards at inference time, where unsafe continuations may still arise. We propose BRT-Align, a reachability-based framework that brings control-theoretic safety tools to LLM inference. BRT-Align models autoregressive generation as a dynamical system in latent space and learn a safety value function via backward reachability, estimating the worst-case evolution of a trajectory. This enables two complementary mechanisms: (1) a runtime monitor that forecasts unsafe completions several tokens in advance, and (2) a least-restrictive steering filter that minimally perturbs latent states to redirect generation away from unsafe regions. Experiments across multiple LLMs and toxicity benchmarks demonstrate that BRT-Align provides more accurate and earlier detection of unsafe continuations than baselines. Moreover, for LLM safety alignment, BRT-Align substantially reduces unsafe generations while preserving sentence diversity and coherence. Qualitative results further highlight emergent alignment properties: BRT-Align consistently produces responses that are less violent, less profane, less offensive, and less politically biased. Together, these findings demonstrate that reachability analysis provides a principled and practical foundation for inference-time LLM safety. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2509_21528 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Preemptive Detection and Steering of LLM Misalignment via Latent Reachability Karnik, Sathwik Bansal, Somil Machine Learning Artificial Intelligence Large language models (LLMs) are now ubiquitous in everyday tools, raising urgent safety concerns about their tendency to generate harmful content. The dominant safety approach -- reinforcement learning from human feedback (RLHF) -- effectively shapes model behavior during training but offers no safeguards at inference time, where unsafe continuations may still arise. We propose BRT-Align, a reachability-based framework that brings control-theoretic safety tools to LLM inference. BRT-Align models autoregressive generation as a dynamical system in latent space and learn a safety value function via backward reachability, estimating the worst-case evolution of a trajectory. This enables two complementary mechanisms: (1) a runtime monitor that forecasts unsafe completions several tokens in advance, and (2) a least-restrictive steering filter that minimally perturbs latent states to redirect generation away from unsafe regions. Experiments across multiple LLMs and toxicity benchmarks demonstrate that BRT-Align provides more accurate and earlier detection of unsafe continuations than baselines. Moreover, for LLM safety alignment, BRT-Align substantially reduces unsafe generations while preserving sentence diversity and coherence. Qualitative results further highlight emergent alignment properties: BRT-Align consistently produces responses that are less violent, less profane, less offensive, and less politically biased. Together, these findings demonstrate that reachability analysis provides a principled and practical foundation for inference-time LLM safety. |
| title | Preemptive Detection and Steering of LLM Misalignment via Latent Reachability |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2509.21528 |