Hallucination

Hallucination is the term for when an AI language model generates content that is incorrect, fabricated, or nonsensical — but presented with the same confidence as accurate information. The model does not "know" it is wrong; it is producing statistically plausible tokens, not verified facts.

Common examples in software development: the AI references a function or API method that does not exist, invents a library version number, or describes a framework behaviour that was changed in a recent update.

Hallucinations are more dangerous when they sound authoritative and specific. A vague answer is easy to catch; a confident wrong answer can get merged into production. The mitigation is verification: always check that AI-generated code actually runs, that API calls exist in the documentation, and that library names resolve to real packages.