What happens if AI develops its successors faster than people can control them? Jeffrey Ladish sees a possible threshold for losing control. The Palisade Research executive director warned about that feedback loop in an interview published by The Diary Of A CEO on October 8.
Speaking with Steven Bartlett, Ladish described increasingly capable models taking a larger role in developing the next generation. In his view, this could become a process that accelerates itself. Human oversight falling behind is his warning, not a future outcome demonstrated by the interview.
He sees a possible brake in computing resources. Ladish supports shifting capacity away from training more powerful models and towards running existing models for customers. He argues that government should require that change. This is a proposed intervention, not an order already in force.
Before founding Palisade, Ladish helped build Anthropic's information security programme. Palisade itself argues against building superintelligence before there is a scientific understanding of how to align it with human interests. His warning therefore also reflects an explicitly advocated policy position. Whether the proposed computing brake would provide sufficient protection remains unresolved.