Glossary · AI & ML

SQL Hallucination

SQL hallucination is when an AI model generates a query that looks correct but references tables or columns that do not exist, or encodes logic that does not match the question asked.

SQL hallucination is the database-specific form of AI hallucination, and it comes in two flavors. The obvious kind fails loudly: the query references "users.signup_date" when the column is actually "created_at", and the database returns an error. The dangerous kind fails silently: the query runs, but it joined the wrong tables or filtered the wrong status values, so the number it returns is wrong while looking authoritative. The fix is grounding–giving the model the real schema as context–plus validating queries before execution and letting humans inspect the SQL. Any AI database tool that hides its generated SQL is asking you to trust hallucinations you cannot check.

Fig — every answer ships with the tables, rows and SQL behind it.