Gen AI, LLMs and new applications of AI have created novel attack surfaces for attackers to probe, often looking to exfiltrate the hoards of sensitive data held in vector data stores. Luckily, the researchers got there first this time and discovered a super interesting attack sequence that tricks CoPilot into locating the most sensitive data it has access to, and then exfiltrating it to the attackers server. Josh, AJ and Vaughan take a look at this new bug in some depth, before looking at other geopolitical cyber developments and threat actor naming conventions.

Support the show