Meta will resume publishing open-weight models and supports allowing model
distillation, reversing a pause after last year’s AI reorganization, Zuckerberg
says. He proposes closer, proactive cooperation between frontier AI labs and
government rather than a fixed 30-day voluntary review window, and urges sharing
intermediate training checkpoints with regulators during model training for
assessment. Meta plans a $1 billion Future Is For Everyone Fund to compensate
communities hosting its data centers and says sustainable infrastructure must
deliver local benefits. Internally, Meta will give its board authority to
approve safety standards and decide whether new models meet those standards, and
Zuckerberg calls for similar board-level and industry-wide governance across the
sector. He reiterates a vision of personal AI agents controlled by users—users,
not developers, should set an agent’s values—and advocates a fully private mode
where Meta or other providers cannot access or grant third-party access to user
data.
