Snowflake says trust and security are essential to AI agents

At Snowflake World Tour London, the company’s Chief Security & Trust Officer, Mayank Uphadyay, said trust and security are central to businesses adopting AI agents. In an interview with TechRadar’s Mike Moore, Uphadyay described two concerns: keeping outside AI systems from breaching a company’s perimeter and ensuring its own AI does not expose data or stray beyond its limits.

Uphadyay said customers are at different stages of AI adoption, but security has become “the enabler, no longer the blocker.” He compared security controls to brakes: they can give drivers confidence, just as guardrails and controls can help customers use AI. Snowflake’s Trust Center connects what he described as three layers, protects customers, and provides visibility into systems and potential improvements.

He said Snowflake has operated for over 12 years and that this history has helped build customer trust. He also highlighted the need to give AI agents appropriate permissions, noting that employees can retain access to systems they no longer need and may run agents with those permissions.

Discussing Snowflake’s Coco and CoWork platforms, Uphadyay described a gradual approach to automation. He said repetitive work could be automated, while people should check the steps between automated tasks and remain involved in mission-critical work. As confidence grows, enterprises can consider automating more, he said.