@zottmann.dev Mike’s summary oversimplifies the tweets. A more accurate reading would be that for Anthropic we assume all the safety is in the contract, and for OpenAI it’s spread over contract and “technical safeguards”.