Who’s Really in Control?
AI agents are stepping into a bigger role inside the enterprise. They are not just providing answers anymore. They are getting work done.
That means interacting with the data and systems businesses depend on, with greater autonomy along the way.
The potential is enormous, but it also raises a critical question. How do you give AI the freedom to work without losing control?
For Snowflake, the answer starts where AI gets much of its value, enterprise data.
Where Data Meets AI
Snowflake’s AI Data Cloud is used by more than 13,900 customers worldwide to build, use, and share data, applications, and AI.
As AI moves closer to enterprise data, Snowflake believes security needs to move with it. The company’s approach centers on giving organizations greater visibility into how AI operates around their data and the ability to govern what happens next.
Building Guardrails for Agents
Before an agent can safely act, an organization needs to know who it is and what it is allowed to do.
That becomes more complicated as agents connect to data and external tools. Snowflake’s Agent Identity addresses that challenge by giving agents verified identities of their own, allowing organizations to set permissions and track their activity.
An agent could be allowed to read from a database, for example, without being allowed to change it.
Bringing Order to AI
Knowing what an agent can access solves one problem. Managing everything it interacts with creates another.
Snowflake’s Cortex AI Gateway provides a central control point for Snowflake and third-party agents as they interact with models, tools, MCP servers, and enterprise systems.
It also provides visibility into agent activity and AI consumption, bringing security, model choice, and cost management together.
Integrations with Aembit, 1Password, Linx Security, Okta, SailPoint, and Saviynt extend that governance across the security ecosystem.
The bigger goal is interoperability without creating a new generation of AI silos. Agents should be able to work across platforms without leaving governance behind.
Security in Layers
Even a properly identified and governed agent can encounter something malicious.
One example is indirect prompt injection, where harmful instructions can be hidden inside data returned through a tool. Jailbreak attempts create another way of pushing AI beyond its intended boundaries.
Snowflake addresses those risks through layered defenses. Snowflake Trust Center helps organizations monitor AI security posture and investigate violations, while AI-native protections are designed to detect threats such as prompt injection and jailbreak attempts.
If one layer misses something, another should still be watching.
Executive Insights
Once those guardrails are in place, the larger question becomes whether enterprises trust AI enough to move it from experimentation into real business operations.
For Mayank Upadhyay, Chief Security and Trust Officer at Snowflake, that is the dividing line.
“The missing piece between AI experimentation and AI operations is trust. Enterprises can experiment with AI without fully understanding how agents behave, but they cannot run mission-critical workflows at scale without knowing who is acting, what those agents can access, and whether every action is governed.”
During our conversation, Upadhyay described agents as naturally exploratory, which can make their behavior harder to predict than traditional software.
Interestingly, AI may also become part of the defense. One model could monitor another for unexpected behavior and intervene when necessary.
I joked that cybersecurity is starting to sound like AI versus AI. Upadhyay agreed. Sometimes, it may take AI to keep AI in line.
From Innovation to Production
Ultimately, these controls matter when companies begin putting agents to work.
For Meltwater, Cortex AI Gateway offers a way to connect agents to the data and tools they need without sacrificing control.
“We see Cortex AI Gateway as a step towards ensuring our agents can securely connect to the right data and tools,” said Aditya Jami, Chief Technology Officer at Meltwater.
At Thomson Reuters, the focus is on maintaining those safeguards as agentic AI moves deeper into professional workflows.
“Organizations need the ability to protect sensitive information and maintain clear controls without limiting innovation,” said Caitlin Halferty, Head of Data & Analytics at Thomson Reuters.
Freedom Needs Trust
The real value of agentic AI comes from letting it do more. That requires enterprises to trust not only what an agent can do, but where it can go, what it can access, and how its actions are governed.
Snowflake is building toward a future where greater AI autonomy does not have to mean giving up control.
The goal is not to hold AI back. It is to know when it is safe to let it work.
Stay Connected
Follow Snowflake on LinkedIn and @Snowflake on X for the latest in AI, data, and security.
#AIDataCloud #AI
About the Author
Angie Apolinar is a Lead Reporter at Cyber Defense Magazine and a Women in Cybersecurity award recipient. She is a graduate student in Cybersecurity and Information Assurance at Western Governors University with a degree in Psychology from California State University, Fullerton. Angie serves as a Cyber Mentor, helping prepare the next generation of cybersecurity professionals. She has also worked on multiple NASA research and workforce development programs, including L’SPACE, where she contributed to mission concepts, systems engineering, software design, and AI-driven aerospace research.
