RAI · Responsible AI Research Lab · University of New Brunswick

— The Lab

The Responsible AI Research Lab (RAI) at the University of New Brunswick, led by Dr. Mahtab Sarvmaili, develops machine learning systems that are trustworthy, interpretable, and accountable. Our work spans the core pillars of responsible AI — explainability, privacy and safety, and security and robustness — and extends into LLM-powered systems and agentic AI, with applications in healthcare, law, and the public interest.

— 01 · Mission

Research that doesn't only work — but can be understood.

Modern machine learning models perform remarkably well, and remain largely opaque. We develop the methods, tools, and frameworks that open these systems up — so that practitioners, regulators, and the people affected by them can see how a prediction was made and decide whether it should be trusted.

— 02 · Foundations

Three pillars of Responsible AI.

Our research is structured around three foundational pillars that together define what it means for an AI system to be responsibly deployed in the real world.

PILLAR I

Explainability, Interpretability & Transparency

Building methods that surface why a model made a prediction — opening the black box so that experts, end users, and auditors can understand, scrutinize, and verify model behaviour.

PILLAR II

Privacy & Safety

Ensuring AI systems handle sensitive data responsibly, do not leak training information, and operate within boundaries that protect individuals and institutions from harm.

PILLAR III

Security & Robustness

Making models resilient against adversarial inputs, distribution shift, and deployment failures — so that systems perform reliably under the conditions of the real world, not just the lab.

— 03 · Research Themes

Where we work.

Four high-level themes anchor our active research programme.

THEME I

Responsible AI

Our overarching research direction — integrating explainability, privacy, safety, security, and robustness into the design and deployment of AI systems.

Read more →

THEME II

Trustworthy AI

Research on reliability, fairness, accountability, and verifiable behaviour — the properties an AI system must demonstrate to earn trust.

Read more →

THEME III

LLM-powered Systems

Applications built on large language models — including retrieval-augmented generation, model editing, alignment, and task-driven control of deployed LLMs.

Read more →

THEME IV

Agentic AI

Building autonomous AI agents that plan, reason over multiple steps, and use tools — with a focus on making them predictable, safe, and aligned with user intent.

Read more →

— 04 · Featured Publications