From Demos to Defensible in Financial Services Copyright & Compliance for Enterprise AI - Naveen Kumar of TD Bank

**Matthew DeMello** (0:17)
Welcome, everyone, to The AI in Business Podcast. I'm Matthew DeMello, editorial director here at Emerge AI Research. Today's guest is Naveen Kumar, head of AI Governance and TD Bank. Naveen joins us on today's program to explore the key challenges slowing AI adoption in banking, from data leakage and prompt injection to shadow AI and hallucinations. Our conversation also covers practical steps like role-based AI guardrails, safe sandboxes for experimentation, hybrid deployments for sensitive data, and treating AI agents as de-risked employees with clear policies and human oversight. But first, are you driving AI transformation at your organization? Or maybe you're guiding critical decisions on AI investments, strategy, or deployment? If so, the AI in Business Podcast wants to hear from you. Each year, Emerge AI Research features hundreds of executive thought leaders, everyone from the CIO of Goldman Sachs to the head of AI at Raytheon and AI pioneers like Yoshua Bengio. With nearly a million annual listeners, AI in Business is the go-to destination for enterprise leaders navigating real-world AI adoption. You don't need to be an engineer or a technical expert to be on the program. If you're involved in AI implementation, decision-making, or strategy within your company, this is your opportunity to share your insights with a global audience of your peers. If you believe you can help other leaders move the needle on AI ROI, visit emerj.com and fill out our thought leader submission form. That's emerj.com and click on be an expert. You can also click the link in the description of today's show on your preferred podcast platform. That's emerj.com/expertone. Again, that's emerj.com/expertone. Without further ado, here's our conversation with Naveen.
Naveen, welcome to the program. It's a great pleasure having you.

**Naveen Kumar** (2:28)
Thank you, Matt. Good to be back here.

**Matthew DeMello** (2:30)
Absolutely. As we were saying a little bit before the microphones turned on, you've moved from retail to financial services. We're hearing from enterprise leaders across financial services, retail, where you used to be as of the last episode, and broader tech spaces that their organizations are accelerating adoption, but copyright licensing and data handling obligations are inescapable in risk workflows with the current status quo. Now that analysts, investigators, and data scientists are increasingly turning to generative tools, the risk is shifting from experimentation to exposure, especially on the output side. So then the challenge, they tell us, becomes this dual mandate. Unlock productivity without creating new copyright liabilities, governance gaps, or regulatory blind spots. Just from the front row, from where you're sitting, where are you seeing copyright and compliance risk usually show up in the actual day to day, and how big is the problem?

**Naveen Kumar** (3:29)
No, that's a great question, Matt. And before we dive into the solution, I think it makes sense to let's unpack where it tends to over copyright and compliance and day to day work, right? So it's sometimes like AI knows too much, right? So copyright risk is when AI output content that it learned from copyrighted resources, for example, internal and external documents, training data, or even online content, right? And so, for example, like a marketer team asked AI to draft report using internal market research. The AI accidentally mirrors phrasing from a copyrighted information. So now you have an issue in hand. So it's like asking a kid to summarize a book and they recite the entire paragraphs before realizing that it's coming from. So it's very tricky in this scenario, right? And compliance risk, you cannot ignore. So when AI produces output that breaks the regulatory, privacy or internal policies, even accidentally, that's a big risk for the firm. We also see data usage and licensing risk. So as we all know, AI needs data to learn and using propriety or license data incorrectly, can create legal headaches. So those are certain things. It's like, again, your AI vendor says, we're training on a publicly available data, but your finance team fed it confidentially, internal reports to fine tune it. Now you have created a copy of propriety data in a model someone else could access.
So that's a big risk for the firm. There's also attributity and, as I say, audibility risk that exists with that, that we are seeing a lot when AI-generated contents, it can be impossible to trace exactly where it came from. So auditors and regulators want to know what is the source. Legal asks for the source of report AI-generated. So your AI says, I don't exactly remember where it came from. That's an issue. So I'm seeing all this quite a bit, like AI leveraging data to create output. There's always an output pressure, but not realizing when is the copyright content being used versus not used.