The Importance of Secure AI Agent Deployment in Enterprises

Article presented by NTT DATA AIVista

A recent report by VentureBeat highlighted a concerning trend in enterprise AI deployment – 69% of companies are still using AI agents that share credentials, leading to higher security risks. At VB Transform 2026, industry experts Mukesh Karki, CTO of NTT DATA AIVista, and Mayank Upadhyay, Chief Security and Trust Officer at Snowflake, emphasized the critical need for secure and scalable autonomous systems.

Karki stressed the importance of not only fixing identity issues but also implementing action-level authorization and tamper-resistant audit trails for every AI agent interaction. According to him, proving compliance in a regulatory environment is essential for enterprises to operate securely.

The Risks of Shared Credentials in AI Security

Upadhyay pointed out that many assumptions from traditional software practices do not apply to AI agents. With AI agents constantly evolving and exploring, shared credentials and static API keys can lead to unintended consequences and security vulnerabilities. The lack of granularity in permissions can expose organizations to significant risks.

Scoped credentials are a basic requirement in industries like insurance, healthcare, and finance. However, Karki emphasized that scoped credentials alone are not sufficient. Enterprises need to establish rules-based and action-based authorization mechanisms tailored to each agent’s specific tasks and jurisdiction.

Challenges in Governance for AI Agents

Unlike human employees, AI agents require specialized governance structures. Karki suggested treating agents like interns – entities that require monitoring and gradual trust-building. Upadhyay outlined a three-layer approach to AI agent governance, covering identity, model, and data layers to ensure proper oversight and control.

Enterprises looking to audit their AI agent governance should focus on addressing static secrets permissions and shadow AI through MCP gateways. Proactive measures like confidence scoring and sandboxing can help mitigate risks associated with autonomous AI actions.

According to Karki, designing systems with provability in mind is crucial for successful AI agent deployment. Retrofitting governance mechanisms post-deployment can be challenging, highlighting the importance of incorporating governance from the ground up.

This article is sponsored content produced by NTT DATA AIVista. For more information on sponsored articles, please contact sales@venturebeat.com.