Chris Olah argues that AI development must not remain exclusive to a few large corporations. He advocates for broader access to the tools and data required for safety research. This push for decentralization aims to prevent Anthropic and its peers from becoming the sole arbiters of model safety and alignment.