Own the end-to-end response to law enforcement and government requests for user data, from intake and validation through production and closeout, including subpoenas, court orders, search warrants, preservation letters, and emergency disclosure requests from domestic and foreign authorities. This research continues many of the directions our team worked on prior to Anthropic, including: GPT-3, Circuit-Based Interpretability, Multimodal Neurons, Scaling Laws, AI & Compute, Concrete Problems in AI Safety, and Learning from Human Preferences.