Trust The Typical

TL;DR

Trust The Typical (T3) introduces a novel safety framework for LLMs that detects unsafe prompts by modeling safe prompt distributions, achieving state-of-the-art results without harmful data training.

Contribution

T3 operationalizes safety as out-of-distribution detection, requiring no harmful examples and generalizing across languages and domains with high efficiency.

Findings

Achieves state-of-the-art performance on 18 safety benchmarks.

Reduces false positives by up to 40x compared to existing methods.

Maintains less than 6% overhead in production settings.

Abstract

Current approaches to LLM safety fundamentally rely on a brittle cat-and-mouse game of identifying and blocking known threats via guardrails. We argue for a fresh approach: robust safety comes not from enumerating what is harmful, but from deeply understanding what is safe. We introduce Trust The Typical (T3), a framework that operationalizes this principle by treating safety as an out-of-distribution (OOD) detection problem. T3 learns the distribution of acceptable prompts in a semantic space and flags any significant deviation as a potential threat. Unlike prior methods, it requires no training on harmful examples, yet achieves state-of-the-art performance across 18 benchmarks spanning toxicity, hate speech, jailbreaking, multilingual harms, and over-refusal, reducing false positive rates by up to 40x relative to specialized safety models. A single model trained only on safe English text transfers effectively to diverse domains and over 14 languages without retraining. Finally, we demonstrate production readiness by integrating a GPU-optimized version into vLLM, enabling continuous guardrailing during token generation with less than 6% overhead even under dense evaluation intervals on large-scale workloads.