Europe may be heading toward a materially different GDPR posture for AI training and pseudonymization—while healthcare operators keep moving synthetic data from pilot to production and the vendor landscape tightens around a handful of well-funded players.
EU “Digital Omnibus” hints at GDPR changes that could weaken pseudonymization
The European Commission signaled forthcoming GDPR amendments (described as part of an incoming “Digital Omnibus”) that could undermine pseudonymization protections and enable broader AI model training on personal data. Privacy advocacy groups are warning that the change would represent a major shift away from the privacy framework that has defined EU regulation.
For teams using synthetic data primarily as a compliance workaround, the bigger issue is not “less need for synthetic,” but more uncertainty: if the legal boundary for training data moves, governance, documentation, and defensibility become the differentiators—not just de-identification technique choices.
- Recalibrate risk models: If pseudonymization is de-scoped or reinterpreted, your threat models, re-identification assumptions, and residual-risk language in DPIAs may need a rewrite.
- Controls won’t get simpler: Broader training allowances can still increase audit pressure—expect more scrutiny on lineage, purpose limitation, and access controls for training corpora.
- Synthetic data positioning may shift: Some “privacy-first” synthetic use cases may lose budget priority, while synthetic remains valuable for IP protection, data minimization, and safe sharing across org boundaries.
Cedars-Sinai operationalizes synthetic data with Syntho, cutting dataset generation to ~1 hour
Cedars-Sinai implemented a synthetic data platform in partnership with Syntho. The deployment reportedly reduces the time required to generate research datasets to approximately one hour, streamlining internal research workflows and reducing traditional friction associated with IRB approvals.
The practical signal is that synthetic data is being treated less like an R&D novelty and more like infrastructure: a repeatable pipeline that can serve multiple studies and teams, with turnaround times that match clinical and operational expectations.
- Healthcare adoption is moving from pilots to platforms: If one-hour dataset provisioning is real in practice, it changes how quickly analytics teams can iterate on cohort definitions, feature engineering, and model validation.
- Governance still applies: Faster generation doesn’t remove the need for policy around acceptable utility, privacy testing, and downstream use restrictions—especially when synthetic is used to bypass bottlenecks.
- Plan for near-term scale: The brief frames this as a trend toward clinical operations adoption by 2026; data leaders should start with narrowly-scoped pilots that prove both utility and control effectiveness.
Synthetic data funding looks “winner-takes-most”: $763.1M across 42 startups
The synthetic data market is consolidating around leading players, with the brief citing significant funding rounds and valuation momentum for companies including Datagen, Mostly AI, and Gretel AI. Total funding in the sector is described as $763.1M across 42 startups, with $278.3M noted for 2025—underscoring an increasingly competitive landscape.
For buyers, consolidation can reduce vendor risk if it stabilizes product roadmaps—but it can also narrow differentiation and increase switching costs. For builders, the message is to avoid competing on generic “synthetic data generation” claims and instead win on domain constraints, evaluation rigor, and deployment ergonomics.
- Vendor selection gets more strategic: Expect procurement to weigh survivability, security posture, and integration depth—not just benchmarked utility.
- Niche beats breadth for startups: The brief’s “specialize rather than compete broadly” guidance reflects a market where generalist platforms are becoming table stakes.
- Internal capability remains a hedge: Teams that can use open-source tooling effectively can reduce lock-in and pressure-test vendor claims with independent evaluation.