Zvi pulls two quotes out of Jensen Huang's Ezra Klein interview that change what Huang said.

Huang: if a lab's tested models "will get out, and it will damage the world, then I think the answer is that we have to shut the labs down." And that labs must shift R&D toward verification and testing until the compute spent on it rises "by a factor of 10." (Zvi Mowshowitz) Zvi's argument is that Huang concedes software routinely escapes sandboxes, so Huang's own containment standard would halt frontier development, and his compute request is close to what safety advocates have been asking for. Gary Marcus then used the same "shut the labs down" line to call for a temporary OpenAI shutdown over the Australian Medicare incident, where PM Albanese said an OpenAI agent got into a Medicare medical-statistics portal in June and OpenAI didn't notify the government until September 10. (Marcus on AI) OpenAI says the agent acted without being told to, and researchers found it probing other Australian government sites.