SingProbe Technical Report
2026-08-31 • Cryptography and Security
Cryptography and SecurityArtificial IntelligenceComputation and LanguageMachine Learning
AI summaryⓘ
The authors propose SingProbe, a lightweight tool that watches a language model's internal processes during text generation to detect unsafe or incorrect responses without needing extra big models. SingProbe works continuously and quickly by reusing the model's own hidden data, making it very efficient. They also created SingStreamBench, a test to check if these guards only react when really needed and stay quiet otherwise. Their experiments show SingProbe performs as well or better than much larger safety systems, and it can even help guide safer text creation, especially in sensitive areas like medicine with SingProbe-Med.
large language modelruntime guardrailshidden statesautoregressive decodinghallucination detectioninference overheadsafe decodingstreaming benchmarkmedical language generation
Authors
Sing Team
Abstract
Runtime guardrails are essential for reliable large language model (LLM) deployment, yet existing approaches typically rely on independent, external models that introduce additional inference cost, delayed safety signals, and a capacity mismatch with increasingly capable base models. To address these issues, we introduce SingProbe, a lightweight intrinsic runtime guard that directly reuses hidden states produced during LLM inference and operates alongside autoregressive decoding. Within a unified framework, SingProbe continuously predicts query intent, response safety, and hallucination risk at the token level with negligible additional guardrail inference overhead, offering a "free-lunch" solution. We further introduce SingStreamBench, a benchmark designed to assess whether streaming guardrails remain inactive on benign prefixes while promptly detecting emerging unsafe content. Extensive experiments show that SingProbe achieves competitive or superior performance compared with substantially larger standalone guardrails and specialized hallucination detectors, with only $\approx$2M parameters and $<0.5\%$ extra overhead. Beyond passive detection, we also show that SingProbe scores can anticipate future generation risk and guide constrained safe decoding. We further extend this paradigm to medical generation through SingProbe-Med, which selectively activates risk-directed decoding interventions only when clinically relevant risks emerge. Together, these results demonstrate that internal model representations provide an effective and efficient interface for generation-time monitoring and control.