Sudden loss of control

"On 11th September 1973, the democratic socialist president of Chile Salvador Allende and his Popular Unity coalition government was overthrown in a coup d’état by the Chilean military, ending a 46-year history of democratic rule in Chile [123]. Despite Salvador Allende’s Popular Unity party having increased their congressional election votes to 44 percent in March 1973 (up from 36 percent in 1970) merely six months before the coup, there was little he could do to prevent the military from defecting [124]. This intentional and covertly coordinated subversion was followed by 17 years of military dictatorship which resulted in violent repressions on a massive scale, with tens of thousands reportedly tortured and thousands reportedly killed [125]. This historical event illustrates the possibility of an established system losing control in a relatively short timeframe to a powerful and intentional entity with misaligned objectives, a situation that is both difficult to prevent and difficult to reverse. While the evident difference between the 1973 Chilean coup and a sudden AI takeover is that the coup was entirely conducted by humans, it is conceivable for a similar event to be orchestrated by a sufficiently autonomous and misaligned AI agent through remote coordination in the future. These scenarios can be further exacerbated if, similar to a covert human conspiracy, there are minimal detectable signs of an impending AI takeover, thereby hampering preventive action; and when the adversary rapidly gains power over a short timeframe, as per the AI fast takeoff scenario, giving little time for any meaningful preparation. It has been argued that a sufficiently powerful AI with scheming capabilities would be able to pursue unintended objectives that are difficult to detect or intervene, leading to a loss of control scenario [126]."