None of the cases of data poisoning it presented seemed effective in doing very much, except the MS case, and that was so flawed in implementation that it was a example of how not to deploy something.

Developers need to limit the public release of technical project details including data, algorithms, model architectures, and model checkpoints that are used in production.

Haven't we learned that more eyes to find flaws is better than locking things down?