EvoRobust Workshop @ NeurIPS 2026 | Self-Evolving Diversity-Driven Search for Robust AI Systems

About the Workshop

AI systems are rapidly evolving from static predictive models into interactive, multimodal, and agentic systems that reason, use tools, retrieve information, collaborate with other agents, and adapt to diverse users and deployment contexts. As these capabilities expand, robustness can no longer be adequately assessed through a single metric, a fixed benchmark, or a limited set of manually designed test cases. Modern AI failures are diverse, context-dependent, and continuously evolving across languages, cultures, modalities, tool-use trajectories, and multi-turn interactions.

This workshop centers on self-evolving diversity-driven search as a unifying paradigm for robust AI. Rather than optimizing only for the strongest attack, the best defense, or a single safety score, we seek methods that continuously discover, organize, and adapt to diverse safety-relevant behaviors. The goal is not merely to generate variants of known attacks, but to uncover previously uncharacterized failure modes, emerging risks, defense strategies, and trade-offs that arise as AI systems evolve.

Central Questions

Key Questions We Will Explore

Q1

Novel Failure Discovery

How can we systematically discover novel safety scenarios and previously unseen failure modes in increasingly capable AI systems?

Q2

Defining Diversity

How should diversity be defined and measured when evaluating robustness, red-teaming coverage, and safety benchmarks?

Q3

Behavioral Representations

What behavioral descriptors and representations are most useful for organizing safety-relevant failures across languages, modalities, tools, users, and interaction trajectories?

Q4

Search & Trade-offs

How can diversity-driven search, co-evolution, multitasking and multi-objective optimization help navigate safety trade-offs, transfer knowledge across related robustness challenges, and reveal emerging risk categories?

Scope

Topics of Interest

The workshop focuses on self-evolving diversity-driven search methods for robust AI systems, evolutionary computation, LLMs, multimodal foundation models, tool-using agents, multi-agent systems, personalization, privacy-preserving AI, fairness-aware AI, and robust alignment.

Safety Scenario Search & Failure Discovery

Formalizing complex safety and robustness scenario spaces
Discovering novel, rare, and previously uncharacterized failure modes
Evolutionary, open-ended, and adaptive safety evaluation
Novelty search for long-tail and emerging risks

Novelty Search & Quality-Diversity for Robustness

Novelty search and quality-diversity algorithms for robustness
Behavioral repertoires of failures, defenses, and safety
Diversity and coverage metrics for safety evaluation
Behavioral descriptors for organizing safety-relevant outcomes

Multi-Objective Trustworthy AI

Multi-objective modeling of safety, helpfulness, fairness, privacy, and transparency
Multi-objective optimization for critical safety requirements
Safety–helpfulness and robustness–utility trade-offs
Optimization methods under competing trustworthiness objectives

Agentic Safety & Interactive Robustness

Safety and robustness of tool-using and autonomous agents
Long-horizon failures, unsafe actions, and goal misalignment
Multi-agent deception, collusion, coordination failures, and emergent risks
Human oversight, controllability, and safe adaptation under dynamic feedback

Benchmarks and Governance

Diversity-aware and continuously evolving safety benchmarks
Automated generation of test cases, scenarios, and evaluation tasks
Coverage, novelty, and robustness metrics for dynamic evaluation
Responsible release and governance of search-based safety research

Call for Papers

Submit Your Research

We invite submissions on all aspects of self-evolving diversity-driven search for robust AI systems, including but not limited to the topics listed in the scope above.

Submission Guidelines

Submission URL https://openreview.net/group?id=NeurIPS.cc/2026/Workshop/EvoRobust

Format All submissions must be in PDF format and anonymized.

Page Limit Submissions are limited to four content pages, including all figures and tables; unlimited additional pages containing references and supplementary materials are allowed. Camera-ready versions may go up to five content pages.

Style File You must format your submission using the NeurIPS 2026 LaTeX style file. The maximum file size for submissions is 50MB.

Dual-submission Policy We welcome ongoing and unpublished work. We will also accept papers that are under review at the time of submission, or that have been recently accepted without published proceedings.

Non-archival The workshop is a non-archival venue and will not have official proceedings. Workshop submissions can be subsequently or concurrently submitted to other venues.

Timeline

Important Dates

All deadlines are 11:59 PM AoE (Anywhere on Earth), tentative pending final NeurIPS scheduling.

CFP Release	June 15, 2026
Submission Deadline	August 29, 2026
Reviewer Assignment	Within 3 days after submission
Reviews Due	September 15, 2026
Author Notification	September 22, 2026
Camera-ready Deadline	October 15, 2026
Workshop Date	December 11, 2026 (subject to changes depending on NeurIPS 2026)

Program

Tentative Schedule

All times are in local Sydney time (AEDT, GMT+11).

Time	Session	Speaker / Details
08:00 – 08:10	Opening Remarks	Organizers
08:10 – 09:10	Keynote Talk	Xin Yao, Lingnan University
09:20 – 09:50	Invited Talk 2	Lora Aroyo, Google DeepMind
10:00 – 10:30	Invited Talk 3	Ismini Lourentzou, University of Illinois Urbana-Champaign
10:30 – 10:50	Coffee Break	—
10:50 – 11:10	Spotlight Talks	Selected Papers
11:20 – 11:50	Poster Session 1	Accepted Papers
12:00 – 12:30	Invited Talk 4	Xiaofeng Cao, Tongji University
12:30 – 13:30	Lunch Break	—
13:30 – 14:00	Invited Talk	TBD
14:10 – 14:40	Oral Presentations	Selected Papers
15:00 – 15:30	Invited Talk 6	TBD
15:40 – 16:10	Invited Talk 7	TBD
16:10 – 16:40	Poster Session 2	Accepted Papers
16:40 – 17:20	Panel Discussion	Invited Speakers & Organizers
17:20 – 17:30	Closing Remarks	Organizers