Submissions

Examples of topics

Adversarial Attacks & Defences: 

• Attacks on AI/Multi-Agent Systems;

• Defences Against Adversarial Attacks; 

• AI Attacks on Humans: Deception, Social Engineering, Disinformation; 

• Backdoor Attacks, Data Poisoning; 

• Strategies for Robustification; 

• Dynamics of Attacker-Defender Co-Evolution across Timescales

Security-by-Design in Multi-Agent Systems: 

• Verification of AI Agents; Secure Mechanism Design;

• Functionality-Security Trade-offs; Coercion-Free, Privacy-Preserving AI Mediators; 

• Security in Human Multi-Robot Systems; 

• Zero-Trust Approaches; 

• Operation Design Domain Specification of Frontier ML Systems; 

• Epistemically Robust Security;

• Accessibility

Limits of Security & Privacy: 

• Formal Limitations & Impossibility Results concerning Privacy-Enhancing Technologies; 

• Detectability in Adversarial Contexts; 

• Verifiability of AI/Multi-Agent Systems; 

• Challenges with Large Language Models; 

• Moral Hazards & Risk Compensation Effects

Ethics, Compliance & Fairness: 

• Detecting Bias & Ensuring Fairness; 

• Preventing Collusion in Marketplaces; 

• Coercion Detection & Prevention; 

• Multi-Agent Systems in Critical Scenarios; 

• Auditing Theory; 

• (Distributed) MASEC Enforcement Mechanisms

Preventing Involuntary Failures: 

• Data Privacy & Security Breaches; 

• Handling Network Congestion & Deadlocks; 

• Misalignment Challenges in the Security Context; 

• Tackling Misinformation; 

• Coordination in AI-AI & Human-AI Cooperation; 

• Mediation failure; 

• Performativity of Algorithms;

Foundational Research in AI Security: 

• Agent-Based Modeling for Security, including (Inverse) Generative Social Science; 

• Multi-Agent Learning & Reinforcement Learning; 

• Automated Mechanism Design; 

• Explainable AI & Causal Inference; 

• Philosophical & Ethical Implications; 

• AI Safety views on multi-agent security; 

• Political Economy of MASEC; 

• Differential Development of MASEC Technologies; 

• Security Dilemma in MASEC: Are purely defensive capabilities possible?

Submission instructions

Submissions must be made via OpenReview. They must be anonymized, up to 4 pages long (excluding references and appendices) and use the NeurIPS 2023 LaTeX template. Appendices can be added to the main PDF.

Each paper will receive at least two reviews; both authors and reviewers will be anonymous throughout the process.

The papers should report original research, provide synthesis of previous works or develop novel environments. Short opinion and review papers are welcomed. Authors can upload concise versions of parallel submissions to other conferences such as NeurIPS main conference or ICLR. We accept dual submission but discourage submitting to multiple NeurIPS workshops.

All accepted papers will be available on the workshop website, but no formal workshop proceedings will be published.