Ex-OpenAI Researcher Advocates for 'Sovereign User Models'

andrew_n_carr · x · 2026-08-26

Jonathan Ward, a former post-training research manager at OpenAI, published a manifesto on building 'sovereign, user-owned models.' Arguing that current AI is bottlenecked by a lack of 'robust goals,' Ward proposes models that are faithful, fully understanding, and controlled by the user. He discusses 'data-maxxing'—converting the continuous stream of a user's digital life into training data to create a proactive 'Great Helper.'

Original post →

More from AGI Musings

AGI Musings channel →