The work is devoted to the problem of unintentional insider, which is a consequence of the user's deviations from normal behavior (deviation) when being in a state of fatigue, stress, affect, prolonged performance of routine tasks, etc.; the relevant prerequisites for threats to information security are substantiated. Using the author's analytical model of such behavior, a method is proposed to counteract unintentional insider by increasing the resistance to user behavior deviation of the instructions themselves. The method consists of 12 following steps: formalization of instructions, isolation of logic and interface elements, selection of features of interface elements, identification of user groups, determination of the level of deviation, building a behavior model, specifying limit values for violations, solving an optimization problem (for correcting instructions), clarifying instructions, adapting instructions, testing instructions, correcting the model. A graphical diagram of the method is presented with its division into the expert's field of action, a decision support system (unified as automation tools), as well as interaction with external components (data, algorithms, and tools). The possibility of using large language models to help an expert perform a number of steps is substantiated.