What happened
LLMs Prove False Mathematical Theorems in MathArena Benchmark.
Why it matters
The MathArena results highlight how frontier LLMs can hallucinate rigorous-sounding proofs for completely false premises when prompted, creating risks for builders relying on models for automated verification or critical reasoning tasks.
Severity rationale
[S3] frontier models fabricated rigorous-sounding proofs for false theorems, but only within a public benchmark with no production harm
Source: matharena.ai
See also: Hallucination Incident Index