AI Policy Watchtower

Logs changes to, and breaches of, the safety policies, model cards and model specifications of OpenAI, Anthropic, Google, xAI, Meta, Microsoft and three smaller developers. It also tracks whether each change was disclosed. This grew out of the project’s Seoul Commitment Tracker (February 2025), which found six of the original sixteen signatories had not met their Seoul commitments, although this has now been surpassed by legislation.