Risk Recognition
Most AI mistakes in management are not random. They fall into four recognizable patterns. Once you can name them, you can catch them before they reach a real employee or a real decision. This is the single most protective lesson in the module, because it turns "AI sometimes gets things wrong" into a specific checklist you can run.
The Framework
The model confidently generates false information, including fake facts, fake citations, and fake quotes. This is common, not rare. Even the best LLMs hallucinate somewhere between 4% and 86% of generated facts depending on the domain and task, with higher rates on novel, niche, or interpersonal topics.
Risk zone: facts, statistics, citations, people-specific claims
You over-rely on the AI's output and accept it without enough independent checking, especially when you are busy. This is a structural feature of attention under load, not a personal weakness. It gets a full lesson in a later module.
Risk zone: every decision made under time pressure
The output systematically skews because the training data was skewed. This shows up most dangerously in people decisions: hiring, performance, promotion. The model can reproduce historical discrimination and present it as neutral analysis.
Risk zone: any decision involving people
The model only knows what you put in front of it, and even then, when the input gets long, it can silently drop earlier parts. You experience this as the AI "forgetting" something you told it earlier in the conversation.
Risk zone: long inputs, complex conversations, multi-part documents
In Practice
Reza's story
Reza, who is skeptical by nature, runs a test. He asks an AI tool to recommend an approach to a team restructure and to back it up with examples. The output is polished and includes a named case study of a company that reorganized exactly this way with great results. Reza tries to find that case study. It does not exist. That is hallucination, the fabricated citation, and it is the moment Reza's skepticism becomes a skill rather than a posture: the answer was useful as a starting frame and worthless as evidence, and he now knows which is which.
Apply It Today
The next time an AI gives you a fact, statistic, or citation you would actually act on, verify one of them against a real source. Notice how often the "source" is vague, unverifiable, or wrong.
Reflect