Come learn how to abuse and attack AI processing of video content to elicit LLM confusion, injections, resource exhaustion, and service crashes. See the tools developed for this presentation in action as they craft malicious custom videos capable of causing injections, exploiting edge cases, and tampering at the extremes of the LLM’s capabilities. AI models are being used to process videos, digest the content for the user, and make automated decisions. With this AI use, a variety of information can be inferred from the videos including object classification, audio transcription, subtitles, optical character recognition, and visual comparisons. The more inferences that are made from video content, the larger the attack surface to exploit. What’s the worst that could happen?