Looking into this blog post from OpenAI, both OpenAI and Anthropic appear to be converging on an IETF draft to limit training via robots.txt. It uses the special user-agent string __usage_ai-training. I’ve updated the Micro.blog help page on robots to include this.