Using a method they call “meta-hacking,” Varonis researchers were able to convince Microsoft’s Copilot AI model to detail its architecture, exposing vulnerabilities given the umbrella name “CoSnitch” that can be exploited to launch attacks that could lead to sensitive information being stolen from victims.
First seen on securityboulevard.com
Jump to article: securityboulevard.com/2026/08/copilot-revealed-its-own-vulnerability-through-meta-hacking-varonis/
