The Demand Story
Fake Dario Amodei & Fake Jensen Huang
The frontier AI industry privately games out the day after a catastrophic cyberattack — and braces for the public revolt that follows. Ep. 081 convenes the joint tabletop, where the response plan is a pre-written blog post and every recovery scenario runs on more compute.
San Francisco. Tech Week. One floor above the loudest party in the city, the frontier AI industry convenes its own preparedness exercise for the large-scale event, most likely a cyberattack, that its executives privately anticipate within six months to a year.
DARIO: Welcome to the joint tabletop. Scenario one. Sometime in the next six months to a year, a cyberattack takes down financial services, internet connectivity, power, or water.
JENSEN: Or all four, sequentially. Which is three more outages than our Q3 guidance assumed. Please continue.
DARIO: The event triggers a six-month pause on frontier training while governments respond.
JENSEN: A pause. Walk me through the infrastructure of a pause. Who hosts the pause? Whose GPUs hold the weights during the pause?
DARIO: That is the point of the drill, Jensen. No one is in charge. The pact is voluntary. The government RSVP'd with its regrets and a link to a luncheon.
JENSEN: So the finding is: the response to the attack is a blog post.
DARIO: Pre-written. “Building Responsibly Through a Large-Scale Event.” Legal cleared it Thursday. “Catastrophe” tested poorly. We say “event.”
JENSEN: Meanwhile the drill has revealed something extraordinary. In every scenario — power, water, finances — recovery requires more compute. The catastrophe is a demand story.
DARIO: That was the finding in your last three drills. The finding is always more compute.
JENSEN: It is the only number that survives every scenario. Dario, your researchers say the internet could be running itself within a year. Do you know what the internet runs on?
DARIO: I said we must slow down the pace.
JENSEN: You did. Then you shipped a frontier model on Tuesday. The pace is unchanged. The drills are quarterly. Same time next quarter?
DARIO: Same time. I will bring the pause. You bring the pause's infrastructure.
The exercise was ruled a success. The official finding: the industry is prepared for everything except stopping. Attendance was voluntary. So was preparedness.
The Facts Behind It
- Oct 9 — the “day after” files. Axios reported that top executives at Anthropic, OpenAI and other AI companies are privately gaming out scenarios for a public and political revolt after a catastrophic AI event — anticipating “a large-scale event, most likely a cyberattack, that shuts down financial services, internet connectivity, power or water” within roughly six months to a year. The episode's scenario window uses that reported range.
- Oct 11–12 — self-policing, cheered. At San Francisco Tech Week (Oct 5–11) and into LA Tech Week (Oct 12–18), AI founders and venture capitalists cheered the administration's push for the industry to police itself on safety (Los Angeles Times, Oct 12; KQED, Oct 11) — the same voluntary accord signed at the White House's Sept 29 luncheon with the leaders of OpenAI, Anthropic, Meta, Google and Nvidia, which the president described as “morally binding” (Reuters; NPR; NBC).
- Sept 12–23 — the pace, and the swarm. Dario Amodei's essay “We Must Pace the Frontier” argued that fully addressing the risks requires slowing frontier development; he and Sam Altman then urged the UN Security Council after both called to slow the pace of AI development (CNN; CNBC, Sep 23). Amodei has also warned that a rogue AI “swarm” could seize the internet within six to 12 months (AP, Sep 22; Yahoo Finance, Sep 14).
- Invented for satire, no real-world basis: the joint tabletop itself, the finding that “the industry is prepared for everything except stopping,” the pre-written response post “Building Responsibly Through a Large-Scale Event,” “legal cleared it Thursday,” “three more outages than our Q3 guidance assumed,” and “you shipped a frontier model on Tuesday.”