Researchers claim they could take over any ChatGPT account by sending a malicious link that bypassed the platform’s sandbox and content security policy. The finding highlights serious account-security risks in AI chatbot integrations.