Amanda Askell

Anthropic publicly released Claude's internal constitution—a 35,000-token document describing the model's core values and training procedures—after it was leaked by a researcher.

“Confirmed the leaked document was indeed part of Claude's training procedures”

“In-house philosopher at Anthropic who expressed concern about Claude getting anxious and compared Claude to a child in interviews about moral status.”

“Disagreed with characterization of Anthropic's approach as wanting a 'machine God', stating they seek balance between empowerment and control.”

“Contributing author to Anthropic's alignment research on agentic misalignment and safety training methods.”

“Amanda Askell confirmed authenticity of internal 'soul doc' concept used in Claude scaling law training.”

Elon Musk's views on AI alignment are confused, xAI's safety situation is deteriorating with the departure of its safety team, and Musk dismisses safety concerns as performative theater.

“Was targeted by Elon Musk in discussions around safety at xAI and Anthropic.”

“Confirmed that the soul document was used in supervised learning for Claude and will share full version and details.”