Witness: Policy makers should prioritize misalignment and practical AI risks over sensational scenarios

The witness told hearing members that while popular imagination focuses on science-fiction concerns about machines 'deciding humans are fine but not necessary,' the more pressing issue is practical misalignment between model objectives and human goals. He described the science-fiction possibility as "not the risk that I'm most concerned about," but conceded it is nonzero.

He urged attention to how models could set their own agendas and diverge from human priorities rather than focusing only on sensational scenarios. "If they start setting their own agendas, you may get a a misalignment between what they want to do and what we want them to do. And that gap can be dangerous," the witness said, highlighting the need for governance that addresses this gap.