“The security risk is the part we take most seriously,” the team said, citing old code and AI exploit risks.