The transition of Artificial Intelligence (AI) from LLMs to autonomous agents, physical AI, foundation models moving beyond pre-training and enhanced focus on reliability and efficiency of AI are some of themes dominating the NeurIPS 2026 workshops lineup.

Other prominent areas include AI for science, world models, multimodal systems, model evaluation, interpretability, AI safety and security, healthcare and efficient AI.

The NeurIPS 2026 conference which is going to take place in Sydney, Australia in December between 6-13 has selected 102 workshops from 454 valid submissions, with 48 workshops to be held in Sydney, 28 in Paris and 26 in Atlanta. The workshops will take place on December 11-12 in Sydney and between December 12-13 in Paris and Atlanta.

AI FrontPage examined all 102 workshops and identified five recurring themes across the programme: agentic AI, foundation models beyond pre-training, physical AI and robotics, AI for science, and trustworthy and efficient AI.

A large number of the accepted workshops focus on AI agents and agentic systems with workshop topics ‘Agentic AI for Biological Discovery’,workshop for autonomous machine learning research, Agentic AI benchmark and application for enterprise tasks, agentic operating systems, agents in the wild, agentic web ecosystems and systems in which AI agents manage other agents- First Workshop on Meta Agents: Managing Agents that Manage Agents.

Foundation models transitioning from pre-training to post-training are another major theme, with workshops examining their use in language, vision, temporal systems, biology, healthcare, physical sciences and robotics. Some of the workshops based on foundation models are Transitioning from Pre-Training to Post-Training, Foundations of LLM Post-Training in Changing Environments, Diffusion Language Models, Linguistic Principles for Foundation Models, Foundation Models for Temporal Systems, AI at Scale for Clinical Impact (ASCI): Cancer Pathology Foundation Models, foundation models for the brain and body.

A number of workshops in physical AI and robotics is also evident across the programme. These include world models in physical AI, robot learning, robot foundation models, embodied intelligence, spatial reasoning and AI systems capable of interacting with and understanding the physical world.

Another major theme is the use of AI in science and healthcare. The conference will also feature workshops focused on trust, safety and evaluation covering topics such as trustworthy AI for good , language model security, child safety in AI, AI privacy, data provenance.

NeurIPS said it received 477 workshop submissions of which 454 were valid, marking an increase from last year. The organiser mentioned they selected the workshops based on factors including scientific merit, broader impact, originality, relevance to NeurIPS, organiser and speaker quality, diversity and logistical considerations.

Also Read: NeurIPS 2026 Position Paper Track Finds Almost 1/3rd Paper Submissions AI Generated