Hijacking AI Memory: Inside Johann Rehberger's ChatGPT Security Breakthrough