TD Banks on Responsible AI Framework as Agentic Systems Emerge
TD has built its approach to artificial intelligence around a Trustworthy AI framework anchored in fairness, explainability, privacy and accountability. The bank now faces a new test: deploying agentic AI systems across back-office workflows while maintaining the same guardrails.
Jesse Cresswell, a researcher leading responsible AI development at Layer 6-TD's machine learning research arm-said the bank tailors responsible AI training across three distinct groups: front-line staff using AI tools, technical developers building models, and executives making deployment decisions.
Training Scales With Responsibility Level
Before branch staff or contact center employees use generative AI virtual assistants, they receive training on safe and efficient use. Model developers learn how to integrate TD's governance framework into their work. At the executive level, 475 TD leaders attended a two-day training session with Columbia Business School last year focused on responsible AI deployment.
The bank's approach reflects a fundamental belief: humans remain responsible for developing AI, deciding how it gets used, and deploying it. Technology itself doesn't earn trust-institutions do.
Guardrails Evolve With Technology
TD's responsible AI training began eight years ago when predictive models dominated the space. The focus then centered on fairness and privacy-ensuring models made fair decisions and handled client data safely.
Generative AI introduced different challenges. A virtual assistant must provide equally complete and accurate answers in English and French. On privacy, developers must understand what proprietary data the model can access and who controls that access.
Before any model reaches users, it clears a series of checks by the Trustworthy AI team and independent validators. Models are evaluated for accuracy, fairness, explainability, privacy and accountability together, not in isolation. Improving one pillar can inadvertently harm another.
After deployment, TD monitors performance continuously to ensure models stay within established values.
Partnerships Inform Implementation
TD is a founding sponsor of the Vector Institute. Over 500 TD colleagues attended Vector workshops and training sessions in 2025 focused on responsible AI. The partnership helps the bank understand how generative AI systems work so it can develop appropriate guardrails.
Columbia Business School training serves a different purpose: it teaches executives why guardrails matter for business outcomes. When leaders understand the business impact, they support guardrail development in their own divisions.
Agentic AI Gets Constrained Rollout
TD designated 2026 "the year of agentic AI," but the bank is approaching the technology cautiously. Rather than giving AI agents broad autonomy, TD constrains them to specific tasks with predefined steps.
"Our AI agents function like a train," Cresswell said. "We're building the tracks by instructing our agents on what they should be doing at every step of a workflow."
The bank starts with lower-risk back-office applications and maintains human review. Employees verify the accuracy of the agent's work before it moves forward. This human-in-the-loop approach acts as another guardrail.
TD positions AI agents as assistants that help colleagues work faster, not as autonomous decision-makers. The distinction matters for both governance and trust.
For executives building AI strategies, AI for Executives & Strategy courses cover responsible governance frameworks and implementation approaches. Those working directly on model development may benefit from Generative AI and LLM Courses that address fairness, privacy and monitoring requirements.
Your membership also unlocks:
