Researchers publish the algorithms. The algorithms wants peer reviewed to cause minimizing in privacy laws, statistically speaking, whatever. The data on there runs the algorithms and publishes only the statistics. Essentially, the algorithm only becomes only another statistical method. Then the data never leaves the hand of data owner, let alone post-processing or synthetic data, or whatever.