AI Governance: Enabling Innovation, Not Just Policing Compliance

The Evolving Role of AI Governance

Xin Tu, a prominent Risk Executive in AI, Data, IT, and Cyber within Financial Services, highlights a critical challenge facing governance teams today: simultaneously responding to immediate regulatory demands while building proactive frameworks for agentic AI. She likens this to "putting the wheels on while the plane is trying to take off." The core issue is that traditional, compliance-heavy governance models are being bypassed by business units eager to innovate with AI. Tu stresses that for AI governance to be effective, it must shift its perception from being a barrier to an enabler of responsible innovation.

Strategies for Effective AI Governance

To achieve this shift, Tu proposes several key strategies:

• Change Perception and Engagement: Governance teams must stop being seen as the function that says "no." Instead, they should actively partner with business units, elaborating on risks, explaining impacts, and offering responsible alternatives rather than shutting down initiatives. This requires a change in both messaging and behavior, fostering collaboration over confrontation.
• Risk-Based Triage and Automation: For high-volume AI applications, human oversight can quickly degrade into complacency. Tu advocates for a risk-based approach, where low-risk use cases can utilize sampled checks, while high-risk scenarios require established mechanisms, automated common checks, and targeted spot-checking. This directs limited human attention to where it matters most.
• "Govern Like a Manager": Drawing a parallel to managing human employees, Tu suggests treating AI systems with similar supervisory rigor. This involves feeding AI models with codes of conduct, defining boundaries, and establishing consequences for deviations, much like managing human behavior within an organization.

Addressing Foundational Challenges

Tu also points out a deeper, often overlooked problem: the neglect of foundational data infrastructure. While enterprises aggressively fund AI initiatives, the "unglamorous work" of cleaning up data governance and addressing technical debt often goes underfunded. She warns that even perfectly running AI algorithms will fail to deliver outcomes if the underlying data quality is poor. Investing in robust data governance is not merely a "nice-to-have" but a fundamental necessity for successful AI adoption and innovation, forming the essential ecosystem that supports AI.