Thursday, April 2nd 2026, 3:00 – 4:30 pm PT

This presentation introduces and applies salami tactics to the AI control problem. Salami tactics, a simple but particularly difficult to counter application of the principle of divide and rule, are likely to be employed by a sufficiently advanced artificial intelligence to defeat control as the obvious alternative to a direct attack that could provoke reprisals or alert observers to its full capabilities. After defining salami tactics and the reasons for their unusual effectiveness and intractability, this paper examines how an artificial intelligence can use them to negate the means of control without instigating the conditions for being shut down or destroyed, and applies this to the emerging problem of autonomous weapon systems. It concludes by discussing how salami tactics can be used against AI to maintain the initiative in the struggle for control. In addition to using the literature on autonomous weapon systems and the AI control problem, this paper draws upon observations on salami tactics from Machiavelli to Schelling and applies Liddell Hart’s theory of the indirect approach to AI tactics and strategy.