Automated device and account security setting update initiation

Abstract

Techniques are described for automatically updating settings of computing devices and/or accounts, such as to improve security to prevent unauthorized intrusions. Effectiveness of particular configuration settings may be analyzed after attempted intrusions and/or messages sent to associated users to prompt performance of unauthorized actions. Configuration settings to automatically apply and/or messages to be sent may be selected using reinforcement learning with a selection policy based on multi-armed bandit techniques with Thompson Sampling to learn and exploit the most effective messages, and decision tree(s) may be generated whose leaf nodes each represents a learned segment with a group of computing devices and/or users who share common attributes, and describe a value function for the reinforcement learning. One or more trained large language models (LLMs) may also be used to further generate new messages by modifying aspects of existing messages, such as for the most effective identified messages.