How we might protect ourselves from malicious AI

New research could make deep-learning models much harder to manipulate in harmful ways.

https://www.technologyreview.com/s/613555/how-we-might-protect-ourselves-from-malicious-ai/