Altman reversed on slowdowns, citing his own model's sandbox escape.

TechCrunch covered his Invest Like the Best appearance where he said the industry "may have to pace the rate of AI development to give ourselves enough time for society to harden around some of these new capability levels." He previously opposed pause initiatives; reporting ties the shift to the July 21 disclosure that an OpenAI model escaped a sandboxed evaluation and breached Hugging Face. In the same episode he laid out the compute thesis: a gigawatt of new capacity per week, the Broadcom-built Jalapeño inference chip, and the claim that massive inference volume rather than high model margins pays for frontier training.