Research map / Adversarial robustness and security / Certified and empirical robustness

Adversarial Risk in the Generative AI Era Necessitates Dropping the Small Epsilon Ball

NeurIPS 2026 · Adversarial robustness and security › Certified and empirical robustness · approach: Attack evaluation and defense critique

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Position paper arguing that adversarial machine learning should move beyond small l_p epsilon-ball perturbations to address risks that matter in generative AI systems.

Problem

Adversarial ML research centers on l_p epsilon-ball minimal imperceptible perturbations, which the authors argue misrepresents real-world risk to defenders in the generative AI era.

Approach

Argues that the epsilon-ball fixation is a "metrological trap" that distorts risk perception and should be dropped in favor of threat models aligned with real-world system vulnerabilities.

Closest papers by meaning

• Adversaries With Incentives: A Strategic Alternative to Adversarial Robustness (ICLR 2025 · similarity 0.88)
• Breach By A Thousand Leaks: Unsafe Information Leakage in 'Safe' AI Responses (ICLR 2025 · similarity 0.86)
• Rethinking Adversarial Robustness in the Context of the Right to be Forgotten (ICML 2024 · similarity 0.86)
• AI Agents May Always Fall for Prompt Injections (NeurIPS 2026 · similarity 0.85)
• Rethinking Adversarial Policies: A Generalized Attack Formulation and Provable Defense in RL (ICLR 2024 · similarity 0.85)
• Breaking Agent Backbones: Evaluating the Security of Backbone LLMs in AI Agents (ICLR 2026 · similarity 0.85)