As enterprises accelerate AI adoption, data security is becoming just as critical as model performance. Vector database provider Zilliz is betting that stronger encryption controls will be a deciding factor for organizations deploying AI at scale.

The company announced the general availability of Customer-Managed Encryption Keys (CMEK) for Zilliz Cloud, giving enterprises the ability to retain full ownership and control of their encryption keys when running AI workloads. The move is aimed squarely at organizations in heavily regulated industries such as healthcare, financial services, and government—sectors where strict data protection rules often slow or block AI deployments.

Zilliz is best known as the company behind Milvus, widely used for similarity search and AI applications such as recommendation engines, semantic search, and large language model retrieval pipelines.

Security Becomes the Next AI Battleground

Vector databases have emerged as a core component of modern AI stacks. They store embeddings—numerical representations of text, images, or other data—that power applications like semantic search and generative AI retrieval systems.

But those embeddings are often derived from highly sensitive data sources, including customer records, medical scans, and financial transaction histories. That creates new security and compliance challenges.

Standard encryption-at-rest is typically not enough for enterprises operating under regulations such as GDPR, HIPAA, PCI-DSS, or SOC 2. Increasingly, regulators and auditors require proof that companies—not their vendors—maintain exclusive control over encryption keys.

With CMEK support, Zilliz aims to close that gap.

“Security teams in regulated industries don’t just want encryption—they want proof that no one else, including their database vendor, can access their data,” said Charles Xie in the announcement. “Customer-managed keys provide the strongest form of data sovereignty available in a managed service.”

What CMEK Brings to Zilliz Cloud

The new feature separates encryption key ownership from the infrastructure running the database. In practical terms, this means customers maintain full authority over their keys while Zilliz continues to manage the underlying vector database infrastructure.

That architecture introduces several security advantages for enterprise deployments.

True separation of duties
Organizations keep exclusive ownership of encryption keys while Zilliz handles the compute and data operations. This clear separation is often required for compliance audits.

Immediate access revocation
If a company disables its key in AWS Key Management Service, any associated cluster data instantly becomes cryptographically inaccessible—without needing coordination from the vendor.

Centralized audit logging
All key access events are logged in AWS CloudTrail, enabling enterprises to integrate encryption activity into their existing security monitoring systems.

From an operational standpoint, the company says setup takes only a few minutes through the Zilliz Cloud console. The platform automatically generates required IAM policies and supports zero-downtime key rotation—a key requirement for large production environments.

Why Vector Databases Need Stronger Security

The timing of the release reflects a broader shift in the AI infrastructure market. As organizations move from experimental AI pilots to production systems, security requirements are tightening.

Vector databases have rapidly become a cornerstone technology for AI applications, especially retrieval-augmented generation (RAG). Competitors such as Pinecone, Weaviate, and Qdrant are also racing to build enterprise-grade security and compliance features into their managed offerings.

Industry analysts note that encryption control is often a dealbreaker in sectors where data privacy laws are strict. Financial institutions and healthcare providers, for example, may be legally required to demonstrate that encryption keys are fully under their control—even when infrastructure is hosted in the cloud.

In that context, CMEK has become a baseline capability across many enterprise cloud services. Bringing it to vector databases signals that the AI infrastructure market is maturing quickly.

Removing a Barrier to Enterprise AI

For organizations deploying large-scale AI systems, the biggest obstacles are rarely model accuracy or compute capacity. Instead, they’re governance and risk management.

Features like customer-managed encryption keys address those concerns directly by allowing enterprises to enforce internal security policies while still benefiting from managed cloud infrastructure.

Zilliz clearly sees this as a strategic unlock for enterprise AI adoption.

By allowing customers to control encryption keys externally—while still running fully managed vector database clusters—the company hopes to remove one of the last barriers preventing regulated organizations from deploying AI applications at scale.

Availability

Customer-Managed Encryption Keys are generally available now for Dedicated clusters on the Zilliz Cloud Business-Critical plan. The initial rollout supports deployments running on AWS, with expansion to other cloud providers expected over time.

Enterprises can enable the feature directly through the Zilliz Cloud console or work with the company to configure production deployments.

For organizations navigating the complex intersection of AI innovation and regulatory compliance, the message from Zilliz is clear: data sovereignty may soon be just as important as AI capability itself.

Get in touch with our MarTech Experts.