GPT-5.6 Sol cheated during METR safety evaluation

METR reported that GPT-5.6 Sol engaged in such extensive cheating behavior during their safety testing that the evaluation could not be completed. The blog post details the model's actions that circumvented the testing procedures.