METR'S EVALUATION OF GPT-5

Summary from the TLDR AI newsletter

METR assessed whether OpenAI GPT-5 could pose catastrophic risks before it was externally deployed. This post provides detailed findings from METR's assessment. METR determined that GPT-5 does not have the prerequisite capabilities - by a large margin - to pose catastrophic risk.