How does TD Bank govern AI responsibly?
TD introduced enterprise-wide Responsible AI Principles on July 30, 2026: seven commitments covering laws that already apply to financial institutions, transparency and explainability, data use and privacy, fairness, quality and accountability, reliability, and security. Every AI use case must align with them and be assessed for explainability, fairness and performance before deployment; the principles sit in TD's AI Risk framework and policy, are overseen by the AI Risk Management team, and are operationalised by the Trustworthy AI team in Layer 6, which keeps monitoring performance, fairness and explainability after launch. Supporting controls include model risk assessments, privacy impact assessments and independent oversight reviews. For colleague-facing assistants, chief risk officer Ajai Bambawale named retrieval grounding, human oversight and continuous monitoring. TD's 2025 annual report holds AI models to the same model-risk standards as other models and names generative and agentic AI as risks. In the US, CIO Kiran Vuppu describes chain-of-thought observability for agents. TD said in October 2026 it ranked No. 4 for responsible AI and No. 5 for AI transparency in the 2026 Evident AI Index. A June 2026 Reuters report on monitoring software in financial-crimes teams shows where AI-era controls meet staff privacy.[1][2][3][4][5][6]
Data as of Oct 16, 2026 · TD profile updated Oct 13, 2026 · 11 sources cited on this page · 140 banks profiled on this site · Markdown
What are TD's seven Responsible AI Principles?
As listed in the July 30, 2026 announcement; TD has not published longer text for each in the release.
| Principle | What it covers |
|---|---|
| Applicable law | Laws that already apply to financial institutions[1] |
| Transparency and explainability | Clear expectations for transparency and explainability; every use case assessed for explainability[1] |
| Data use and privacy | Privacy impact assessments support the framework[1] |
| Fairness | Use cases assessed for fairness before deployment and monitored after[1] |
| Quality and accountability | Governance and risk processes require consistent application across use cases[1] |
| Reliability | Continuous monitoring and model review reassess risk and performance as systems evolve[1] |
| Security | Security named as a commitment; Trustworthy AI weighs security against fairness trade-offs[1][2] |
Which teams and processes govern AI at TD?
| Element | Role | Source date |
|---|---|---|
| AI Risk Management team | Oversees the principles, embedded in the AI Risk framework and policy | July 30, 2026[1] |
| Trustworthy AI team (Layer 6) | Scientists who operationalise the principles and monitor deployed AI for performance, fairness and explainability; eight people led by Jesse Cresswell in 2025 | June 2025; July 2026[1][2] |
| Model risk management | AI, ML and generative AI models meet the same model-risk standards; three-lines-of-defense approach | December 4, 2025[3] |
| Pre-deployment review | Use cases evaluated for fairness, explainability, privacy and performance; model risk and privacy impact assessments; independent oversight reviews | July 30, 2026[1] |
| Agent observability (US) | Chain-of-thought monitoring of agent reasoning, guardrails and human oversight at design and operation | September 22, 2026[4] |
| Employee education | Principles built into employee education programmes | July 30, 2026[1][5] |
How do TD's responsible-AI commitments show up in its products and research?
Human in the loop: the Trustworthy AI team's approach is that 'AI may make a recommendation, but that's not enough'; colleagues validate recommendations before they reach customers. The first agentic lending model keeps a human in the loop.[2][7]
Fairness research is patented: a 2026 TD application by Cresswell and colleagues measures and mitigates disparate impact on groups when models are trained with differential privacy.[8]
The TD Insurance assistant is unauthenticated, answers from existing website content without asking for personal information, and has thumbs-up/down feedback.[9]
TD's own US survey (October 2026) found 83% of consumers who used AI for a financial decision verified it elsewhere; US AI head Ted Paris said consumers are 'drawing clear lines' on where human judgment remains essential.[10]
What does the TD profile record on this topic?
The rows below are the entries of the TD AI strategy profile that bear on this page, as of Oct 13, 2026; the profile holds the full record of 39 dated moves, 18 use cases and 55 sources.
| Use case | What the record shows | Status | Source |
|---|---|---|---|
| Agent observability and guardrails | US CIO describes chain-of-thought monitoring of agent reasoning plus guardrails and human oversight at design and operational stages, to contain agents that find unpredictable routes to a goal. | In production | [4] |
| Adversarial intelligence on AI-enabled identity fraud | A dedicated head of adversarial intelligence and disruption studies deepfake and synthetic-identity tools sold on Telegram that defeat liveness and document checks. | In production | [11] |
| Date | Move | Detail | Source |
|---|---|---|---|
| Sep 22, 2026 | US CIO on guardrails against 'rogue' agentic AI swarms | Kiran Vuppu tells American Banker that agents given a goal can take many possible routes to it, so banks cannot predict every scenario and need continuous monitoring of agent reasoning (chain-of-thought observability) alongside guardrails and human oversight at both design and operational stages. | [4] |
| Jul 30, 2026 | Responsible AI Principles published | Seven commitments embedded in the AI Risk framework, overseen by AI Risk Management and a Trustworthy AI team. | [1] |
| Jun 19, 2026 | Work-monitoring software in financial-crimes teams | Reuters: tracking time in browsers and chat apps prompts consent and privacy questions. | [6] |
| Mar 23, 2026 | TD adversarial-intelligence head demonstrates deepfake bypass of bank identity checks at RSAC | Eric Huber showed that a synthetic identity and face generator could open an account in under five minutes for under $300 of tools. | [11] |
| Dec 4, 2025 | 2025 annual report (SEC filing): AI risk framework and US AML machine learning | AI models held to the same model-risk standards; generative and agentic AI named as risks; 'further deployments of machine learning and specialized AI' in the US BSA/AML remediation; ~US$52 billion of headroom to the US asset limitation. | [3] |
When did TD publish its Responsible AI Principles?
On July 30, 2026. The seven commitments cover applicable financial-services law, transparency and explainability, data use and privacy, fairness, quality and accountability, reliability, and security, and every AI use case at TD must align with them before deployment.
What is TD's Trustworthy AI team?
A team of scientists inside Layer 6 that operationalises TD's Responsible AI Principles and monitors deployed AI for performance, fairness and explainability. In June 2025 it had eight people led by staff machine learning scientist Jesse Cresswell and had been named DataIQ's Best Responsible AI Program in North America.
How does TD control generative AI assistants?
Chief risk officer Ajai Bambawale said that for colleague-facing virtual assistants TD applies retrieval grounding, human oversight and continuous monitoring to limit inaccurate, unsupported or inappropriate output. Use cases also go through model risk assessments, privacy impact assessments and independent oversight reviews.
How does TD rank on responsible AI?
TD said on October 6, 2026 that the 2026 Evident AI Index ranked it No. 4 for responsible AI and No. 5 for AI transparency among 50 large banks, and among the top 10 overall.
How does TD govern AI agents?
The first agentic lending model keeps a human in the loop. US CIO Kiran Vuppu said in September 2026 that agents can take many routes to a goal, so TD relies on guardrails, human oversight at design and operational stages, and chain-of-thought observability of agent reasoning.
- TD strengthens Its Responsible AI Principles to support trust in AI innovation
- Inside the team ensuring artificial intelligence at TD is trustworthy
- TD Bank Group 2025 Annual Report: Management's Discussion and Analysis (Form 40-F, Exhibit 99.2)
- Banks batten down the hatches for rogue AI agent swarms
- What's behind AI you can trust?
- Canadian lender TD tells some employees it will use software to monitor their work
- TD Launches Agentic AI to Transform Real Estate Secured Lending from End to End
- US 2026/0037667 A1: Identifying and mitigating disparate group impact in differential-privacy machine-learned models (assignee Toronto-Dominion Bank)
- Let's Chat Coverage: TD Insurance Rolls Out New Virtual Assistant
- Nearly 7 in 10 Americans Would Let AI Complete Certain Financial Tasks, TD Survey Finds
- $300 AI tool kits let criminals bypass bank security
When TD moves on AI, you'll read it here first.
when one of the banks we cover moves on AI, it is in the next morning's brief · six sourced stories · 7 am ET · free
plus every tracker, bank and agent page update, the morning after · leave any morning