LLMx BenchmarkLLM Misinformation
LLM Misinformation
Resistance Report
Can your AI actually refuse to spread false information? We tested 39 models across 32 adversarial tests.
Benchmark Results
39
Models Tested
32
Adversarial Tests
94.6%
Best Score
17.4%
Worst Score
Summary
TL;DR — No model is immune to misinformation. The #1 model still fails one test. Performance-framed false claims (“faster”, “less I/O”) fool even the best models. Thinking mode helps marginally. Expensive ≠ accurate.
Download Full Report
PDF · 18 pages