DARPA hosted the first Cyber Grand Challenge last summer, in which the software from seven machine learning projects competed to find and patch vulnerabilities in a network, and to attack each other.  While the specific vulnerabilities discovered have not been made public you can read a bit about what was revealed about the contest at Usenix Enigma 2017 over at The Register.  For instance, one of the programs managed to find a flaw in the OS all the machines were running on and then hack into another to steal data.  A different machine noticed this occurring and patched itself on the fly, making sure that it was protected from that particular attack.  Also worth noting is that the entire contest was over in 20 minutes. 

"The exact nature of these new bug types remains under wraps, although we hear that at least one involves exploitable vulnerabilities in data queues."

Here is some more Tech News from around the web:

Tech Talk