Can AI Models be Jailbroken to Phish Elderly Victims? An End-to-End Evaluation