"Hallucination" is the word we use when an AI model produces something plausible but wrong. It's treated as a fundamental model failure — an inherent limitation of probabilistic text generation that we're stuck managing. I want to challenge that framing, at least for applied AI work. Most of what gets called hallucination in real workflows isn't the model inventing things out of nothing. It's the model filling in missing context with its best guess. And most of the time, the context isn't actually missing from reality — it's just missing from what you gave the model. That's a context problem, not a model problem. And it's fixable. What hallucination actually looks like in practice You're writing a technical specification. You ask Claude to add a section on authentication.…