New theory certifies risk for automated security decisions

A new paper introduces a decision-contract theory to address the challenge of certifying risk for automated security decisions, particularly in the context of LLM-based intrusion detection systems. The theory aims to prevent systems from appearing risk-bound simply by abstaining from action. The research proposes an error-conservation law and a non-degenerate actionability certificate to ensure that error is accounted for across automation, human deferral, and semantic masking. Empirical results on three intrusion detection datasets using six different LLMs demonstrate that the proposed method can achieve target risk levels in over 90% of configurations while maintaining a high rate of correct automation. AI

IMPACT Introduces a theoretical framework to improve the reliability and trustworthiness of automated security decisions made by AI systems.

RANK_REASON The cluster contains a single academic paper published on arXiv. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• alphaXiv
• arXiv
• CatalyzeX
• DagsHub
• Gotit.pub
• Hugging Face
• Mitre ATT&CK
• ScienceCast

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →