Anthropic on fully automated AI research:

these models need much stronger safety standards, so its push to slow frontier AI was largely driven by the belief that they could be trained soon