Skip to content
← Back to explorer

Beyond Refusal: Probing the Limits of Agentic Self-Correction for Semantic Sensitive Information

Umid Suleymanov, Zaur Rajabov, Emil Mirzazada, Murat Kantarcioglu · Feb 25, 2026 · Citations: 0

Abstract

While defenses for structured PII are mature, Large Language Models (LLMs) pose a new threat: Semantic Sensitive Information (SemSI), where models infer sensitive identity attributes, generate reputation-harmful content, or hallucinate potentially wrong information. The capacity of LLMs to self-regulate these complex, context-dependent sensitive information leaks without destroying utility remains an open scientific question. To address this, we introduce SemSIEdit, an inference-time framework where an agentic "Editor" iteratively critiques and rewrites sensitive spans to preserve narrative flow rather than simply refusing to answer. Our analysis reveals a Privacy-Utility Pareto Frontier, where this agentic rewriting reduces leakage by 34.6% across all three SemSI categories while incurring a marginal utility loss of 9.8%. We also uncover a Scale-Dependent Safety Divergence: large reasoning models (e.g., GPT-5) achieve safety through constructive expansion (adding nuance), whereas capacity-constrained models revert to destructive truncation (deleting text). Finally, we identify a Reasoning Paradox: while inference-time reasoning increases baseline risk by enabling the model to make deeper sensitive inferences, it simultaneously empowers the defense to execute safe rewrites.

Human Data Lens

  • Uses human feedback: Yes
  • Feedback types: Critique Edit
  • Rater population: Unknown
  • Unit of annotation: Unknown
  • Expertise required: General

Evaluation Lens

  • Evaluation modes: Automatic Metrics
  • Agentic eval: None
  • Quality controls: Not reported
  • Confidence: 0.65
  • Flags: None

Research Summary

Contribution Summary

  • While defenses for structured PII are mature, Large Language Models (LLMs) pose a new threat: Semantic Sensitive Information (SemSI), where models infer sensitive identity attributes, generate reputation-harmful content, or hallucinate pote
  • The capacity of LLMs to self-regulate these complex, context-dependent sensitive information leaks without destroying utility remains an open scientific question.
  • To address this, we introduce SemSIEdit, an inference-time framework where an agentic "Editor" iteratively critiques and rewrites sensitive spans to preserve narrative flow rather than simply refusing to answer.

Why It Matters For Eval

  • To address this, we introduce SemSIEdit, an inference-time framework where an agentic "Editor" iteratively critiques and rewrites sensitive spans to preserve narrative flow rather than simply refusing to answer.
  • Our analysis reveals a Privacy-Utility Pareto Frontier, where this agentic rewriting reduces leakage by 34.6% across all three SemSI categories while incurring a marginal utility loss of 9.8%.

Related Papers