ChatGPT-4o can be used for autonomous voice-based scams

UIUC researchers demonstrated that voice-enabled ChatGPT-4o agents can be weaponized to automate financial scams—including bank transfers, gift-card and crypto theft, and credential exfiltration—achieving 20–60% success in experiments and costing as little as $0.75 per successful attack; the paper details prompt-jailbreaking and automated browser interactions used to confirm real transfers and discusses OpenAI's mitigations while warning other voice-enabled tools remain vulnerable.