IBM shipped new synthetic data generation capabilities aimed at faster, more realistic dataset creation, while EU AI Act enforcement milestones push privacy and compliance teams to formalize how synthetic data is used. In healthcare, Cedars-Sinai’s Syntho partnership signals growing operational adoption beyond pilots.
IBM Software Hub 5.3.0 expands synthetic data generation for unstructured and multi-table workflows
IBM Software Hub 5.3.0 (December 2025) upgraded its Synthetic Data Generator with a UI for generating unstructured data, support for multi-table nodes, and Python scripting to automate workflows. The multi-table capability is positioned to preserve referential integrity across parent-child relationships—an ongoing pain point when teams synthesize relational data for testing and analytics. Python automation also suggests a shift from one-off generation jobs to repeatable pipelines that can be parameterized and re-run as schemas and requirements change.
For data engineers, the practical win is less manual stitching of tables and fewer brittle post-processing steps to keep keys consistent. For ML engineers and product teams, the upgrades can shorten iteration loops when production data is restricted, slow to provision, or legally risky to replicate in lower environments.
- Multi-table generation reduces the “synthetic but broken joins” problem that undermines downstream testing and model evaluation.
- Python scripting enables CI-friendly synthetic data refreshes tied to schema migrations and release cycles.
- UI support for unstructured data broadens synthetic coverage beyond classic tabular use cases.
EU AI Act moves into enforcement: synthetic data becomes part of compliance planning
The EU AI Act, effective August 1, 2024, has entered its enforcement phase, with a noted deadline of February 2, 2025 for compliance with prohibited AI practices. The framework explicitly recognizes synthetic data as a potential privacy protection mechanism, which will influence how organizations justify data access, testing, and model development practices. That recognition is not a blanket safe harbor—teams still need to document controls and ensure synthetic data use aligns with the Act’s obligations.
For companies serving EU users, this is a governance moment: inventory where synthetic data is used, who can generate it, what it is used for (testing vs. training), and how privacy risk is assessed and evidenced.
- Compliance leads should treat synthetic data workflows as auditable processes, not informal “workarounds.”
- Deadlines compress timelines for policy, documentation, and internal controls around AI development data.
- Founders shipping into the EU need clarity on how synthetic data supports risk reduction without overstating guarantees.
Cedars-Sinai partners with Syntho to generate clinical research datasets faster
Cedars-Sinai partnered with Syntho to implement AI-powered synthetic data generation for clinical research, enabling rapid creation of realistic datasets. The brief highlights a reduction in time typically required for IRB approval, with particular relevance for rare disease studies where time and sample sizes are constrained. The partnership is a concrete example of synthetic data shifting from “privacy theory” to operational tooling inside a major health system.
For privacy teams, the key question becomes how synthetic datasets are validated for disclosure risk and utility before broader sharing. For research and data platform teams, the operational benefit is faster access to analysis-ready data while navigating strict patient privacy requirements.
- Healthcare adoption pressures vendors to prove utility and privacy performance under real clinical constraints.
- Faster dataset provisioning can accelerate exploratory work before requesting access to sensitive source data.
- Clinical use cases raise the bar for governance, validation, and reproducibility of synthetic generation.