Enterprises using multiple AI models are underestimating failure rates by 2.25x

Businesses relying on multiple AI models to cover each other's weaknesses are significantly underestimating failure rates by a factor of 2.25, according to a study analyzing 67 advanced models from 21 providers. The "co-failure ceiling" reveals that combining these models doesn't create the safety net expected, as they often fail on overlapping prompts. This insight is crucial for companies to avoid overconfidence in their AI systems, as it highlights the need for more robust error-checking and fail-safes to ensure reliability.