A selective use of test results to make AI rankings look better.
It is like showing only your best test scores. The bad math quiz is suddenly “lost.”
It can trick people choosing AI tools. It can also fool investors. The ranking stops showing real skill.
Leaderboard
It uses cherry-picked scores to mislead the ranking.
Benchmark contamination
Leaked test data can create falsely high scores.
Benchmark Validity
Manipulation makes rankings fail to show real AI skill.