It seems to me the researchers efforts would be better directed toward securing the communication rather than developing an AI tool to guess when that communication is malicious. We've had decent PKI systems since the mid-1990s. It shouldn't be a heavy lift to ensure that communication with unmanned vehicles has mutual authentication, confidentiality, integrity, and non-repudiation. I'm actually aghast that perhaps these things aren't being done already.
While it's glossed over in the story, the researchers indicated a false positive of 2%. Think about that. You have a drone about to carry out a strike, and the instructions come to abort. One out of every 50 missions, though, this algorithm might ignore the command, interpreting it as malicious.
If you want better military command and control in the civilized world, simple: have every government adopt a law that the vanguard of any attack must include the president, czar, minister, congress people, representatives, etc. that voted to take that action. But continuing to distance those individuals from the consequences of their decisions by clouding the picture with more technology is a bad idea.
