Teradata adds privacy & resiliency for enterprise AI

Teradata has introduced new privacy, security and resiliency features for enterprise AI data protection. The additions cover data privacy, access controls and recovery measures across its cloud data platform.

The changes are aimed at organisations seeking to protect sensitive information used in AI systems while keeping critical operations available and recoverable if systems are disrupted.

The launch includes Poly-Anonymization, which reduces exposure of personally identifiable information by replacing PII fields with stable identifiers, and format-preserving encryption, which protects sensitive values without changing their structure in downstream systems. Teradata also added post-quantum protections for data in transit through TLS 1.3 hybrid-key exchange, immutable backup protection, and database-level high availability based on real-time replication.

The privacy functions are delivered through an OEM partnership with Anonomatic. Its PII Vault stores underlying identity data separately from the analytical environment. Authorised users can restore identity where required, while others work with data that preserves analytical relationships without broadly exposing who it relates to.

Privacy focus

Teradata argues that AI is changing enterprise data protection because systems increasingly move sensitive information across users, applications, models and software agents. This makes perimeter-based security less effective on its own and raises the risk that non-identifiable data could be re-identified when AI systems combine multiple datasets.

The new controls are designed to sit closer to the data layer. They include native dynamic data masking, data classification and tagging, column-level encryption, and a global identity framework across Teradata's cloud environments.

Teradata has also expanded policy-based access management through a partnership with Immuta. The company said this adds dynamic attribute-based access control, allowing data policies to be applied consistently across users and AI workloads without rewriting governance rules or duplicating datasets.

"Privacy, security, and resiliency have historically been treated as separate problems with separate solutions. That approach doesn't hold up when AI is moving sensitive data across users, applications, models, and agents simultaneously. This announcement reflects how we think those capabilities need to work together, and it's the foundation for how we intend to extend that protection as AI environments grow more complex and the actors operating within them multiply," said Sumeet Arora, Chief Product Officer at Teradata.

Security changes

Post-quantum readiness is another part of the update. Teradata said TLS 1.3 hybrid-key exchange combines classical and post-quantum cryptographic algorithms in a single handshake for data in transit, as companies assess the longer-term risks posed by advances in quantum computing.

Teradata also said Anonomatic PII Vault is built as a post-quantum-secure environment with multiple security layers intended to protect identity information held outside the main analytical workflow.

Teradata cited internal testing at large scale that it said showed the Anonomatic-based privacy controls could support enterprise analytical workloads without the performance penalties often associated with privacy technologies. According to the company, multi-table join queries on protected data used fewer Teradata resources than equivalent queries on unprotected data, although results would vary depending on workload and configuration.

"Enterprise AI creates a fundamentally different privacy challenge. Organisations need to use sensitive, individual-level data to build effective AI, but the traditional tools for protecting that data weren't designed to preserve what AI needs to work. Our partnership with Teradata makes it possible to reduce identity exposure at enterprise scale without forcing organisations to choose between protection and utility," said Fleck.

Recovery layer

Alongside privacy and access controls, Teradata has introduced new resiliency measures under what it calls Enterprise Data Protection. The approach aligns protection levels with the criticality of specific workloads rather than applying the same architecture to every system.

Database-level high availability has been added as another layer for critical workloads, using real-time replication for continuous synchronisation and low recovery targets. Immutable backup protection is intended to ensure recovery points cannot be modified, encrypted or deleted, including by privileged administrators, as a measure to reduce ransomware risk.

Teradata said these measures are intended to help restore not only data after an incident, but also the business context and operational information that AI systems rely on to resume important processes.

Privacy features delivered with Anonomatic are available on Teradata Cloud across AWS, Microsoft Azure and Google Cloud. Native security features, including dynamic data masking, data classification and tagging, and column-level encryption, are also available across those cloud environments, with some functions additionally offered for Teradata Factory and on-premises deployments. Immutable backup protection is available on Teradata Cloud across all three hyperscale cloud platforms.