Limits of Statistical Assurance for Embodied and Agentic AI under Adaptive Adversaries
cs.CR, cs.LG, stat.ML
Submitted: 2026-10-13
Updated: 2026-10-13
Comments: 66 pages, 10 figures
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Terminology
Sources
- On Evaluating Adversarial Robustness
- AgentPoison: Red-teaming LLM Agents via Poisoning Memory or Knowledge Bases
- AgentDojo: A Dynamic Environment to Evaluate Prompt Injection Attacks and Defenses for LLM Agents
- Alignment faking in large language models
- Not what you've signed up for: Compromising Real-World LLM-Integrated Applications with Indirect Prompt Injection
- Time-uniform, nonparametric, nonasymptotic confidence sequences
- Steering Evaluation-Aware Language Models to Act Like They Are Deployed
- Propagating Unsafe Actions in LLM Controlled Multi-Robot Collaboration via Single Robot Compromise
- Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training
- Adversarial Attacks on Robotic Vision Language Action Models
- Evaluation Awareness in Language Models Has Limited Effect on Behaviour
- The Shawshank Redemption of Embodied AI: Understanding and Benchmarking Indirect Environmental Jailbreaks
- AttackVLA: Benchmarking Adversarial and Backdoor Attacks on Vision-Language-Action Models
- When Robots Obey the Patch: Universal Transferable Patch Attacks on Vision-Language-Action Models
- RoCo: Dialectic Multi-Robot Collaboration with Large Language Models
- Large Language Models Often Know When They Are Being Evaluated
- Probing and Steering Evaluation Awareness of Language Models
- Visual Adversarial Examples Jailbreak Aligned Large Language Models
- Game-theoretic statistics and safe anytime-valid inference
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