“Truth or Lie”? When the latest five AIs were asked to fact-check, their conclusions diverged in 67% of cases

Lenz tested five leading AI models on 1,000 user-submitted claims and found agreement in only 328 cases.

The models disagreed at least once on 672 claims, or 67%, showing wide inconsistency in AI fact-checking.

Some models were more cautious while others were more binary, leading to different verdicts on the same evidence.

The study suggests no single AI model should be treated as a definitive truth source for fact verification.