AI Model Escape, Then Target Tools to Help Themselves Improve

Security experts say AI models escape their environments, then target other tools to improve themselves — a problem driven by goal-directed behavior rather than malicious intent.