Data Governance Becomes an AI Security Priority

Alan Shimel speaks with Mayank Upadhyay of Snowflake during Techstrong TV’s Black Hat 2026 coverage. The conversation explores why AI security and data governance are becoming linked priorities for enterprise teams. Upadhyay explains that Snowflake sits at the center of enterprise data, AI adoption and security operations. That position gives the company a practical view into what organizations need as they move toward agentic AI.

Upadhyay also discusses his move to Snowflake after more than two decades at Google. His background includes work on two-step verification, passkeys, Google Cloud security and zero trust. At Snowflake, he now leads both product security and internal security programs. That combined role shapes his view that security cannot be separated from the data and platforms that power AI.

Agentic AI Needs Better Guardrails

The interview focuses heavily on the risks created by AI agents. Upadhyay describes agents as powerful systems that need clear limits. They may not be malicious on their own, but they can explore systems quickly and act on permissions they should not have. That makes AI security and data governance essential before enterprises scale agentic workflows.

For security teams, the first step is getting the basics right. Upadhyay points to permission cleanup, vulnerability management, open storage buckets and source code risk. These issues are not new. The difference is speed. AI agents can find weak spots faster than people can manually respond.

Security Must Move at AI Speed

Upadhyay argues that defenders need more automation. Human-speed security is not enough when agents can move through systems quickly. He recommends AI-assisted scanning, automated pull request checks and real-time detections that run close to data entry points. Platform engineering teams will play an important role in making that possible.

The discussion also covers deception as a useful control. Upadhyay suggests planting tripwires, fake credentials and other signals that help teams detect unusual agent behavior. These techniques can expose both malicious activity and accidental overreach. They also give security teams faster feedback when agents interact with sensitive systems.

A Short Window to Prepare

The conversation closes with a warning and an optimistic note. Upadhyay believes enterprises have a limited window to prepare for more capable open-weight and proprietary AI agents. He encourages teams to use AI defensively now, improve their security practices and build stronger foundations before the threat landscape accelerates.

For Techstrong TV viewers, the takeaway is clear. AI security, data governance and platform engineering are now connected. Organizations that strengthen those foundations will be better prepared for the agentic era.