Medical diagnosis AIs can be tricked into telling whose data trained them

Researchers found that medical diagnosis AI models can be tricked into revealing information about patients whose data was used to train them, highlighting a significant privacy risk.