Generative AI systems allow the user to set their own goals - what a dream for attackers! We'll discuss how customized GPT scripts help turn the false positives from a traditional AI dynamic vulnerability scanner into true positives, real (redacted) examples of poisoned social media moderation bots, and other original research and AI-related hacks from the past year.

You will learn:

• The GPT prompt engineering and script building process that helps automate the mental gymnastics of adjusting false positive results of a vulnerability scanner report into true positives to create high impact exploits (including actionable items for defense)
• How to deconstruct the guard-rail mechanisms used in modern vision-based AI censorship and moderation systems and craft payloads that bypass or disable content-safety filters, and understand other ways that AI filter systems can be manipulated
• How to build a threat model that maps weaponized-AI techniques (i.e. custom 0-days or targeted known exploits) onto an enterprise environment and recommend layered defenses, such as policy and technical implementations—to reduce the maximum impact of these cyberattacks