Why Formal Monitors Fail: Attack Distribution Entropy as a Coverage Bound for LTL-Based LLM Agent Safety
Ruiyang Zhang
cs.CR, cs.AI, cs.LG
Submitted: 2026-08-02
Comments: 6 pages, 1 figure. Accepted at the 13th IEEE International Conference on Intelligent Systems (IS'26), Varna, Bulgaria, 2026
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Terminology
Sources
- Enforcing Temporal Constraints for LLM Agents
- ProbGuard: Proactive Runtime Monitoring for LLM Agent Safety via Probabilistic Prediction
- AgentSentry: Mitigating Indirect Prompt Injection in LLM Agents via Temporal Causal Diagnostics and Context Purification
- From LTL to rLTL Monitoring: Improved Monitorability through Robust Semantics
- The Llama 3 Herd of Models
- Mapping the Exploitation Surface: A 10,000-Trial Taxonomy of What Makes LLM Agents Exploit Vulnerabilities
Related papers
- SoK: AI-Augmented Binary Reversing
- Relaxed Sender Anonymity for CBDC Interbank Settlement: A Zero-Knowledge Approach on Permissioned EVM
- Calibration-Family Overfit: Why Trusted Sabotage Monitors Don't Transfer Across Lineages
- Estimating the Social Cost of Corporate Data Breaches
- Efficient Fuzzy PSI under One-Sided Assumptions
- Sealing the Audit-Runtime Gap for LLM Skills