IRB Loophole via Synthetic Data

Hospitals and universities are training generative models on real patient records, then using the models’ synthetic outputs to run studies without Institutional Review Board approval. They argue the outputs are not human data, even though training used identifiable sources, promising faster research and easier data sharing. This blurs the line between human‑subjects research and model‑mediated datasets, risking uneven safeguards across institutions. — If synthetic data lets researchers bypass ethics review, regulators must redefine when consent and oversight apply (e.g., at model training) to protect privacy without stalling science.