Why it matters
NDC AI talk on manipulating AI systems and separating theatrical jailbreak demos from the real risk created by deployed workflows. It belongs in the library as red-team framing for how persuasion, prompt pressure, user intent, and system boundaries interact once AI is placed in products.
My takeaway: Red-team work should move beyond clever prompts and ask what the product lets the model read, decide, and trigger. The risk is in the full workflow around the model.