The World Economic Forum is pushing synthetic data policy away from one-size-fits-all rules and toward context-aware governance, with stronger quality benchmarks and explicit safeguards when synthetic data is used for model training.
WEF urges tailored governance and stronger benchmarks for synthetic data
The World Economic Forum (WEF) published a synthetic data report that focuses on how organizations and regulators should evaluate and govern generated datasets. The report recommends more rigorous quality assessment, highlights hybrid training approaches (mixing synthetic and real data), and flags the risk of “model collapse” when synthetic data is used in ways that degrade downstream model performance.
On the policy side, WEF argues regulators should distinguish between different types of synthetic data and adopt context-aware standards rather than treating all synthetic data as equivalent. The report positions synthetic data governance as a multi-dimensional problem—privacy, fairness, and deployment risk—where the right controls depend on the use case, the generation method, and how the data will be used (for sharing, analytics, or training).
- Expect more nuanced compliance asks: If regulators follow WEF’s framing, teams may need to document the “type” of synthetic data used and justify controls based on context (use case, method, and risk), not just claim de-identification.
- Benchmarks become procurement and audit artifacts: Stronger quality assessment guidance raises the bar for vendor evaluation—data leads should anticipate requests for measurable utility, bias/fairness checks, and privacy risk evidence as part of approvals.
- Training pipelines need guardrails: The explicit warning on model collapse reinforces the need for hybrid strategies and monitoring when synthetic data is used in training, especially where feedback loops can amplify errors over time.