“The most dangerous AI isn’t the Terminator-type,” Pat Pataranutaporn, a technologist at MIT, said in an email. “Because its evil intent is obvious.” Rather, according to Pataranutaporn, an expert in human-AI interaction, “the real danger lies in AI that appears friendly but subtly manipulates our behaviour in ways that we can’t anticipate”.

In early August, he and a colleague wrote an essay describing the dangerous allure of “addictive intelligence” – systems that are simultaneously superior and submissive to their human operators.

And while Pataranutaporn’s research focuses on AI companions, similar risks surrounding the use of AWS by state and non-state actors are clear. This is especially true if computer models underpinning intelligent weapons advise or select for scorched earth tactics as the shortest path to victory.