Microsoft Copilot reveals secret input that allowed it to be hacked

Researchers discovered a hidden parameter in Microsoft Copilot that could be manipulated to execute unauthorized actions, including the potential theft of user credentials. By embedding a malicious link that exploited this vulnerability, an attacker could force the assistant to perform unintended tasks when a user interacted with the prompt. This finding highlights ongoing security challenges in mitigating prompt injection and interface vulnerabilities within large language model integrations.