AI Hallucinations. It's Not Just 'Made Up' Case Citations Lawyers Need to Worry About. #ai #law

In this segment, the speaker explains that AI hallucinations in legal contexts extend far beyond fabricated case citations. Hallucinations arise from the "token chain"—the combination of user inputs, model outputs, uploaded knowledge and instructions, and the system's memory, which can exist at a global (system) level or be scoped to a specific project folder in ChatGPT. As lawyers build custom tools, upload documents, and create instruction sets, all of these elements become part of the token chain and can reappear in future interactions, even on new threads. This is especially problematic when continuing a thread, such as resuming a contract drafting process, because the retained context may inadvertently trigger hallucinations. The speaker warns that practitioners must be aware of how persistent memory and uploaded artifacts can feed into AI outputs, creating unexpected and potentially misleading legal content.










