As AI systems become embedded in every product and workflow, attackers are shifting their focus from individual prompt injections to scalable exploitation strategies. This talk reframes LLM security from the attacker’s perspective - how malicious actors think, where they look for broad-scope opportunities, and what makes an AI system an attractive, repeatable target. We will explore the emerging patterns that mirror classic internet-era exploitation - from poisoning the content supply chain to manipulating discovery mechanisms at scale. Attendees will leave with a new mental model for anticipating large-scale AI attacks and understanding why the next generation of threats will not be handcrafted, but industrialized.