Perforce Delphix launches AI-native synthetic data tool for dev teams
Perforce Software has launched Delphix Synthetic Data, an AI-native product designed to let developers, testers and AI agents generate realistic, scenario-specific test data without relying on restricted or incomplete production datasets. The product is part of the existing Delphix DevOps Data Platform and became available on 22 September 2026.
The core proposition addresses a well-documented friction point in modern software delivery: teams building or testing new features often cannot access live production data for privacy, compliance or availability reasons, yet legacy synthetic data tools have been slow, manually intensive and poor at preserving the relationships between data across systems. Perforce cited its own 2026 survey of 518 enterprise technology leaders, which found that only 34% of respondents who had evaluated existing solutions said those tools provided referential integrity, and only 36% said they delivered adequate data realism.
How the product works
Delphix Synthetic Data uses an embedded AI model to scan metadata and database schemas automatically, identifying data structures, entity relationships and business context across sources. It then applies statistical analysis to understand the shape and distribution of the underlying data before generating synthetic equivalents. Users can further refine or extend the generated dataset using natural language prompts, which Perforce says reduces the time from data request to usable test data from days or weeks to minutes.
A notable architectural decision is the bring-your-own-LLM model: customers can run their preferred large language model within their own infrastructure, and the built-in AI component analyses only metadata rather than actual record values. Perforce says sensitive or personal data is therefore never transmitted to the AI layer. The product is accessible through a user interface, REST APIs and Model Context Protocol workflows, the last of which is aimed at agentic development pipelines.
Ilker Taskaya, Field CTO at Perforce Delphix, argued that masked production data is inherently backward-looking. "Testing needs the cases that aren't in production yet, and it needs them to hold together across every database and file format in the environment," he said, adding that the new product generates data "in the shape teams specify, with referential integrity intact."
Jim Mercer, Programme Vice President for Software Development, DevOps and DevSecOps at IDC, described the category requirement plainly: organisations need test data that can be generated quickly, scale across complex environments and satisfy data-privacy requirements.
Market context and competitive landscape
Synthetic data generation has attracted a growing number of specialist vendors, including Mostly AI, Gretel.ai and Tonic.ai, alongside data-masking incumbents that have added generative capabilities. The entry of a platform vendor such as Perforce, which already provides data masking and data delivery through the Delphix platform, positions the product as a consolidation play: buyers already using Delphix for masking can extend to generation without adding a separate vendor.
The broader regulatory environment reinforces demand. GDPR and its national equivalents, the EU AI Act's requirements around training and testing data quality, and sector-specific frameworks such as HIPAA and PCI-DSS all restrict the use of raw production data in non-production environments. As agentic development workflows proliferate, the volume of test data required per sprint is expanding significantly, making manual synthetic-data configuration increasingly untenable.
Perforce has not disclosed pricing, availability by region, or named any launch customers. The company reports customers across more than 80 countries, including a claimed majority of the Fortune 100, but provided no customer-specific validation of the new product in this release. Benchmark data comparing Delphix Synthetic Data against named competitors is also absent. Buyers evaluating the product will want to see independent referential-integrity tests across heterogeneous database environments before committing at scale.