TRUSTMORE 2026

Overview

Modern multimodal AI agents are rapidly moving from research prototypes to production systems. These agents increasingly support high-impact applications in healthcare, finance, legal services, cybersecurity, scientific discovery, and government. However, deployment in regulated and safety-critical environments demands far more than benchmark performance. Production AI systems must operate reliably at scale while satisfying stringent requirements for robustness, auditability, explainability, privacy, observability, and regulatory compliance.

TRUSTMORE 2026 brings together researchers, practitioners, and industry leaders working on the systems, infrastructure, evaluation methodologies, and engineering practices that enable trustworthy AI deployment. We welcome contributions spanning deployed applications, engineering innovations, evaluation frameworks, benchmarks, datasets, and operational experiences that advance multimodal AI agents from research prototypes to production-ready systems.

The workshop places particular emphasis on real-world deployments, scalable infrastructure, practical engineering challenges, and lessons learned from production environments. We especially encourage submissions from both academia and industry that bridge advances in AI research with the realities of deploying multimodal agents safely, reliably, and at scale.

Topics of Interest

We invite original research papers, system papers, demos, benchmarks, datasets, and work-in-progress submissions addressing the challenges of building and deploying trustworthy multimodal agents in real-world environments. We are particularly interested in work that demonstrates practical impact, addresses production-scale challenges, and bridges advances in AI research with the realities of deployment.

1. Demos of Deployed Systems and Practical Applications

This theme focuses on AI systems that operate at scale in production today, emphasizing real-world performance and practical experience over benchmark novelty. We seek end-to-end system papers and live demonstrations that provide insights practitioners can learn from and adopt.

• End-to-End Deployed Agentic Systems: Full-system architectures covering the lifecycle of multimodal production applications, from data ingestion and agent orchestration to monitoring and iterative improvement.
• Multi-Agent Coordination and Autonomous Reasoning: Coordination architectures for agents collaborating across tools, memory, and reasoning steps, including production failure modes and orchestration strategies.
• Practical Applications in Regulated Domains: Multimodal agent deployments in healthcare, finance, legal services, cybersecurity, government, and other high-impact or regulated domains.
• Latency-Aware and Resource-Constrained Deployment: Techniques for meeting constraints on latency, compute, cost, and availability while maintaining reliability, safety, and compliance.
• Production Architectures and Deployment Case Studies: Engineering approaches, system designs, operational lessons, and failure analyses from real-world deployments.
• Human-Agent Collaboration: Human-in-the-loop systems, human oversight, escalation mechanisms, and approaches for effectively integrating autonomous agents into existing workflows.

2. Trustworthy Infrastructure & Methods

This theme focuses on the engineering foundations that make trustworthiness operational at production scale. We welcome work addressing the infrastructure, methods, and runtime mechanisms required to build reliable, observable, auditable, and safe multimodal agent systems.

• Retrieval, Grounding, and Knowledge Integration: Methods and architectures for efficient retrieval, grounding, and knowledge integration to improve the reliability of multimodal agents, including long-context settings.
• Observability and Outlier Detection: Monitoring, logging, runtime diagnostics, and techniques for detecting anomalous or unexpected agent behavior.
• Error Detection and Failure Recovery: Methods for identifying, preventing, containing, and recovering from errors and failures in multimodal agent systems.
• Robust Orchestration and Agent Infrastructure: Scalable orchestration frameworks and architectures for reliable multi-agent and tool-using systems.
• Safety Mechanisms for Autonomous Agents: Runtime safeguards, guardrails, intervention mechanisms, and other approaches for managing risks associated with increasingly autonomous systems.
• Auditability, Compliance, and Governance: Infrastructure and methods supporting traceability, explainability, auditability, regulatory compliance, and responsible deployment.
• Security, Privacy, and Resilience: Secure deployment, access control, privacy-preserving methods, fault tolerance, and resilience for production AI systems.

3. Benchmarks, Datasets & Objective Evaluation

Trustworthy multimodal agents require evaluation infrastructure that goes beyond static accuracy benchmarks. This theme seeks contributions that develop auditable, domain-grounded benchmarks and evaluation methodologies capable of identifying and quantifying failure modes that emerge in real-world deployment.

We are particularly interested in work that treats benchmark construction as a data engineering problem, where dataset sourcing, annotation, provenance, versioning, validation, and coverage are as important as the metrics themselves.

• Production-Oriented and Domain-Specific Benchmarks: Benchmarks designed around real-world tasks and deployment conditions, with attention to dataset provenance, edge-case coverage, and reproducibility across institutional settings.
• Safety and Reliability Evaluation: Evaluation protocols for assessing robustness, calibration, uncertainty, selective abstention, output consistency, and other behavioral properties of multimodal agents.
• Long-Context and Multimodal Evaluation: Methods for evaluating agent behavior across long-running interactions, complex multimodal inputs, and realistic production workloads.
• Continuous Evaluation: Evaluation pipelines and methodologies that support ongoing assessment of deployed agents as models, data, tools, and environments evolve.
• Adversarial Evaluation, Red Teaming, and Jailbreaks: Systematic methodologies for discovering vulnerabilities, policy violations, unexpected behaviors, and failure modes that standard benchmarks may not capture.
• Formal Auditing and Objective Evaluation: Formal or systematic approaches for auditing agent behavior, including evaluation of goal misgeneralization, specification gaming, and other undesirable behaviors.
• Dataset Quality and Reproducibility: Methods for ensuring annotation reliability, label consistency, dataset quality, provenance, and reproducibility of evaluation results.

Submission Types

We welcome the following submission categories:

Submissions should follow the IEEE BigData 2026 template.

Submissions should present original work that has not been published elsewhere or simultaneously submitted to another venue, consistent with IEEE publication policies.

Publication

A key goal of TRUSTMORE 2026 is to provide authors with a formal publication venue for practical and impactful work on trustworthy multimodal agents.

All papers accepted for workshops will be included in the Workshop Proceedings published by the IEEE Computer Society Press, made available at the Conference.

This provides an opportunity for both academic researchers and industry practitioners to share research findings, production experiences, system designs, benchmarks, and lessons learned with the broader AI community.

Why Submit?

TRUSTMORE 2026 provides a dedicated venue for addressing the practical challenges of deploying trustworthy multimodal AI systems in real-world environments.

The workshop will feature:

• Invited keynote speakers from academia and industry
• Technical paper presentations
• System demonstrations and posters
• Lightning talks
• Panel discussions on trustworthy AI deployment
• Networking opportunities between researchers and practitioners
• Opportunities to connect academic research with real-world production challenges

We particularly encourage submissions describing deployed systems, engineering innovations, infrastructure, benchmarks, datasets, evaluation methodologies, operational experiences, and interdisciplinary collaborations.

Work describing practical challenges, negative results, failure analyses, and lessons learned from deploying AI systems is also strongly encouraged.

Review Process

All submissions will undergo double-blind peer review.

Each submission will be evaluated based on:

• Technical quality
• Originality
• Significance
• Practical impact
• Relevance to trustworthy multimodal agents
• Clarity and reproducibility

Authors are encouraged to include anonymous supplementary material, code, datasets, or other resources where appropriate.

The organizers anticipate approximately 1–1.5 months for the review process. The submission deadline may be extended if additional submissions are needed to support a strong workshop program.

Important Dates

All deadlines are Anywhere on Earth (AoE) unless otherwise specified.

The workshop is co-located with IEEE BigData 2026, which is scheduled for December 14–17, 2026. The exact date of the TRUSTMORE workshop will be announced once finalized.

The timeline is designed to provide accepted authors with sufficient time to make travel and visa arrangements before the conference.

Workshop Format

TRUSTMORE 2026 is planned as a full-day workshop and will include:

• Invited keynote talks
• Oral paper presentations
• Poster and demo sessions
• Lightning talks
• Panel discussions
• Interactive discussions between researchers and practitioners
• Hybrid participation, with both in-person and remote participation

The workshop is expected to take place during the December 14–17, 2026 IEEE BigData conference period. The specific workshop date will be announced once confirmed.

Accepted presentations and workshop materials may be made available online following the event.

Organizers

• Manas Gaur, University of Maryland, Baltimore County (UMBC)
• Manas Paldhe, Infinitus Systems
• Nikita Mishra, Amazon
• Youngseo Son, Infinitus Systems

Submission Portal

Submissions will be handled through OpenReview:

TRUSTMORE 2026 OpenReview Submission Portal

Workshop Website

Coming soon

Join Us

As multimodal AI agents become increasingly capable and autonomous, ensuring that these systems are reliable, scalable, safe, and accountable in real-world environments has become a central challenge for the AI community.

TRUSTMORE 2026 aims to bring together researchers and practitioners working at the intersection of AI, systems, infrastructure, safety, evaluation, and real-world deployment.

We particularly welcome contributions that move beyond benchmark performance to address the challenges that arise when AI agents are deployed in production: unexpected failures, monitoring, evaluation, governance, scalability, security, human oversight, and the need to continuously establish trust in increasingly autonomous systems.

We look forward to your submissions and to building a community around trustworthy, production-ready multimodal agents.