A setup where one model's output is passed to one or more additional model instances prompted to challenge, critique, or find errors in it before a final answer is produced.

Borrowed from courtroom drama by AI reliability vendors to make a second prompt asking "check this for mistakes" sound like a hostile witness being broken down under oath.

• Courtroom Theater: Dressing a review prompt in the language of adversarial legal proceedings to imply rigor that was never established.
• Shared Blind Spot: The cross-examiner is usually the same model with a different instruction, so it inherits the same training and misses the same errors it is supposed to catch.
• House-Cleaning Reality Check: Like reading your own shopping list a second time, missing the milk again, and calling the second pass "cross-examination."

A Few Good Men (1990s) scene where Lieutenant Kaffee (Tom Cruise) breaks Colonel Jessup (Jack Nicholson) on the stand through sustained adversarial questioning from a genuinely opposing side.

A second model prompted to look for errors in the first model's answer before it is finalized.

“Our multiagent cross-examiner adversarially validates every generated finding prior to release.”

“We send the first draft back through the same model with instructions to look for mistakes in it.”