[NeurIPS Workshop CFP] Multi-Agent Security Workshop
Swapneel Mehta
Dear colleagues,
We’re glad to announce the first Multi-Agent Security Workshop at NeurIPS 2023 is now accepting submissions!
Important dates:
Submission deadline: September 25, 2023
Acceptance Notification: October 27, 2023
Workshop date: December 16, in-person @NeurIPS 2023,
New Orleans, LA, USA
Workshop Description:
Security is the foundation of cooperation among AI agents, humans, institutions, and nations. Progress in areas such as LLMs/GenAI, multi-agent learning, and human-AI interaction offer great potential, but also highlight new security challenges. Yet, the cyber security, AI capabilities, safety and policy communities have limited platforms to share cross-cutting knowledge. The field of Multi-Agent Security (MASEC) aims to design a blueprint of the emerging and amplified security challenges of our present and future multi-agent world.
At this in-person NeurIPS workshop, we will gather researchers from cyber security, and AI, including AI safety and AI governance, to connect the fields that complement each other. We will showcase on-going research via contributed talks and poster sessions, and provide networking opportunities for research and mentorship discussions (see website for detailed program).
Submission Formats
We welcome 4-page submissions of technical, benchmark and opinion
papers on any topics related to the MASEC themes across disciplines (machine learning, cyber security, cognitive sciences, philosophy, psychology, and more). Our ultimate objective revolves around the development of an ambitious security blueprint for AI cooperation.
We want to think beyond the conventional boundaries of cyber security to encompass not only data privacy and information security, but also individual security—freedom from threats and coercion—alongside corporate and even national security considerations. When we refer to "multi-agent systems", we encompass intricate collaborations among AI agents, humans, institutions, and the underlying cooperative platform infrastructure and rules of engagement. This spectrum covers everything from AI agents in the physical realm, such as autonomous driving systems, to the realm of digital entities like digital assistants, and even Mixed Reality.
Examples of topics include:
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.
For any questions, email us at mas...@googlegroups.com or reach us on
Twitter at https://twitter.com/masecworkshop!
Cheers,
The MASEC @ NeurIPS 2023 Organizing Committee