1st Workshop: Assessing and Improving the Capabilities and Safety of Agents

This workshop brings together two critical threads in the development of autonomous agents: their growing capabilities and the imperative to ensure their safe and aligned deployment. As agents become increasingly capable of long-horizon reasoning, goal-directed behavior, and autonomous decision-making, understanding how to measure and extend these capabilities remains a central research challenge. At the same time, these advances raise urgent questions about robustness, misalignment, risk mitigation, and unintended behaviors.

The workshop will explore current bottlenecks to agent development—including the challenge of collecting meaningful agentic data (via human interaction or synthetic generation) and ensuring generalization without mode collapse—as well as cutting-edge approaches to building agents that are safe by design. By bringing together researchers working on both the capabilities and safety of agents, this workshop aims to foster cross-pollination between communities and develop a shared research agenda for the next generation of autonomous systems.

Aishwarya Agrawal (IVADO, Université de Montréal, Mila), Prithviraj “Raj” Ammanabrolu (UCSD), Andrea Bajcsy (Carnegie Mellon University), Yoshua Bengio (IVADO, Université de Montréal, Mila), Yonathan Bisk (CMU), Joyce Chai (University of Michigan), Alexandre Drouin (ServiceNow & Université Laval), Daniel Fried (Carnegie Mellon University), Zhijing Jin (University of Toronto), Parisa Kordjamshidi (Michigan State University), Bo Li (UIUC), Graham Neubig (Carnegie Mellon University), Siva Reddy (IVADO, McGill University, Mila), Abhik Roychoudhury (National University of Singapore), Ruslan Salakhutdinov (Carnegie Mellon University), Dawn Song  (UC Berkeley), Yu Su (The Ohio State University), Alane Suhr (UC Berkeley), Huan Sun (The Ohio State University), Ivan Titov (University of Edinburgh), Xin Eric Wang (UCSB, Simular), Tao Yu (The University of Hong Kong), Victor Zhong (University of Waterloo, Vector Institute).

FRIDAY, OCTOBER 3^RD, 2025 – A.335 Amphithéâtre Rachelle et Alain-Paris et Foyer

9:00 – 9:20 a.m.: Welcome and Check-In

9:20– 9:30 a.m.: Welcome Address

9:30 – 10:15 a.m.: Towards Internet-Scale Training of Web-Agents
Ruslan Salakhutdinov (Carnegie Mellon University)

10:15 – 11 a.m.: Computer Use: Modern Moravec’s Paradox
Yu Su (The Ohio State University)

11 a.m. – 11:30 p.m.: Coffee Break

11:30 a.m. – 12:15 p.m.: Trusted Automated Software Engineering via LLM Agents:  Experience from AutoCodeRover
Abhik Roychoudhury (National University of Singapore)

12:15 – 2:00 p.m.: Lunch Break (on your own)

2:00 – 2:45 p.m.: Title TBA
Parisa Kordjamshidi (Michigan State University)

2:45 – 3:152 p.m.: Communicating Efficiently with Agents by Reasoning About Informativity
Daniel Fried (Carnegie Mellon University)

3:15 – 3:45 p.m.: Interactive Language Feedback in Guiding Agent Behaviors
Joyce Chai (University of Michigan)

3:45 – 4:15 p.m.: Title TBA
Yoav Artzi (Cornell University)

4:15 – 5:30 p.m.: Social Hour / Mixer

SATURDAY, OCTOBER 4TH, 2025– A.335 Amphithéâtre Rachelle et Alain-Paris et Foyer

9:00 – 9:30 a.m.: Welcome and Check-In

9:30 – 10:15 a.m.: Lessons from the Trenches on Building Usable Coding Agents
Graham Neubig (Carnegie Mellon University)

10:15 – 11:00 a.m.: Multimodal Agents for Solving Complex Reasoning Tasks
Aishwarya Agrawal (IVADO, Université de Montréal, Mila)

11:00 – 11:30 a.m.: Coffee Break

11:30 a.m. – 12:15 p.m. : Spatial and Semantic Awareness in Embodied LMs
Yonatan Bisk (Carnegie Mellon University)

12:15 – 2:00 p.m.: Lunch Break (on your own)

2:00 – 2:45 p.m.: Trust and Control in LLMs: from Faithful Reasoning to Privacy-Preserving Collaboration
Ivan Titov (University of Edinburgh)

2:45 – 3:30 p.m.: Title TBA
Zhijing Jin (University of Toronto)

3:30 – 4:00 p.m.: Coffee Break

4:00 – 4:45 p.m.: Catching the Uncaught: From Safety Risks in Computer-Use Agents to Generalization Failures of Transformers
Huan Sun (The Ohio State University)

SUNDAY, OCTOBER 5TH, 2025 – A.335 Amphithéâtre Rachelle et Alain-Paris et Foyer

9:00 – 9:30 a.m.: Welcome and Check-In

9:30 – 10:15 a.m.: Open Foundations for Computer-Use Agents
Tao Yu (The University of Hong Kong)

10:15 – 11:00 a.m.: Building AI Agents that Reason and Act Like Humans
Xin Eric Wang (UCSB)

11:00 – 11:30 a.m.: Coffee Break

11:30 a.m. – 12:30 p.m.: Panel – The Future of Autonomous Agents
With Joyce Chai (University of Michigan), Tao Yu (University of Hong Kong), Parisa Kordjamshidi (Michigan State University)

12:30 – 2:00 p.m.: Lunch Break (on your own)

2:00 – 2:45 p.m.: Agents for Knowledge Work – Progress and Key Challenges
Alexandre Drouin (Service Now, Université Laval)

2:45 – 3:30 p.m.: From Refusal to Recovery: A Control-Theoretic Perspective on Agentic Guardrails
Andrea Bajcsy (Carnegie Mellon University)

3:30 – 4:00 p.m.: Coffee Break

4:00 – 4:45 p.m.: Guarding the Future: Red Teaming, Alignment, and Knowledge-Enabled Guardrails for Agentic Systems
Bo Li (University of Illinois Urbana-Champaign)

MONDAY, OCTOBER 6TH, 2025 – A.463 GardaWorld and A.469 Espace Stéphen-Crétier

9:00 – 9:30 a.m.: Welcome and Check-In

9:30 – 10:15 a.m.: Towards Building Safe and Secure Agentic AI
Dawn Song (UC Berkeley)

10:15 – 10:45 a.m.: Navigating the Safety-Capability Spectrum when Teaching Agents with Feedback
Prithviraj Ammanabrolu (UCSD)

10:45 – 11:15 a.m.: Coffee Break

11:15 a.m. – 12:00 p.m.: Avoiding Catastrophic Risks from Uncontrolled AI Agency with the Scientist AI
Yoshua Bengio (IVADO, Université de Montréal, Mila)

12:00 – 12:10 p.m.: Closing Remarks