Trump's Super Intelligence Summit, AI Safety Accord, GDP Beats, Midterm Predictions

The episode opens not on policy but on hardware. David Sacks and the hosts spend the first minutes on Starlink, and Sacks offers a technical claim that frames the rest of the hour: a sufficiently dense constellation of satellites that both emits and receives radio waves functions as a phased-array sensor looking down at the Earth. In his telling, the same network that carries the audio can pick out objects that are designed to be invisible from the ground.

It's a stealth airplane detector.

The exchange is comic — someone asks whether they can skip the show's science segment — but it sets up a recurring theme: capability arriving from infrastructure nobody built for that purpose. The hosts then pivot to self-deprecation about the show's own name, joking that a program called This Week in AI somehow never covers AI, before settling on the framing that the four of them are, in effect, four superintelligences discussing the week.

Jason Calacanis sets the timeline: a few weeks earlier the group had been in Los Angeles debating whether frontier AI should be audited or slowed, and then Trump convened a summit in Washington. The guest list spanned the stack — Elon Musk, Jensen Huang, Mark Zuckerberg, Dario Amodei, Sundar Pichai, Satya Nadella — plus the administration. Sacks is credited with shaping the event, and he is asked to explain both the speed and the output.

the only person who could get all the CEOs of all these companies in one room

Sacks's explanation for the speed is personal rather than procedural: only the president could compel CEOs of data-center, chip, hyperscaler and frontier-model companies to sit together, including executives who dislike each other and in some cases are litigating against each other. He calls the result the Bretton Woods of superintelligence and stresses that it was the first time the whole stack was represented at once — a claim about scope, not about the stringency of what was signed.

Sacks describes the agreement as having a layered structure. First, the companies accept that they are the responsible actors — he explicitly rejects deflecting responsibility onto the models themselves, onto anthropomorphizing language, or onto the United Nations. Second, they commit to internal controls verified by an internal team. Third, and this is the part he emphasizes, the process does not stop at what he calls self-policing.

External auditors validate whether the internal controls actually work, and each company sets up an independent committee of its board to receive those reports. Sacks cites Gavin Baker's observation from that morning: once an external auditor's report reaches the board, ignoring it is not an option. The compliance chain therefore runs from engineering controls to auditor to board, rather than from regulator to company.

the governance that follows from it is not voluntary

The enforcement backstop is existing authority rather than new statute: the FTC and SEC can act on public commitments. Sacks argues this means implementation can begin immediately, without legislation and without an international agreement he considers unattainable. His stated goal is public confidence in safe development, and he frames the absence of a Congressional or treaty process as a feature — speed — rather than a gap.

The hosts press on the apparent contradiction of a voluntary agreement with binding consequences. Sacks's answer runs through directors-and-officers liability insurance: a board that receives an adverse audit finding and does nothing exposes itself, and its D&O coverage, to risk. Chamath Palihapitiya restates the mechanism as directors facing personal and insurance consequences, and Sacks agrees the arrangement hooks deeply into board governance.

It has deep, deep hooks into the board of directors.

Later, when asked which provision carries the most force, Sacks names external audits without hesitation. That answer implicitly ranks the audit requirement above the statement of responsibility and above the working group on safety practices, which he says were the pre-existing elements. The novel additions, in his account, were internal controls, external audits and the board committee.

Who performs those audits is left open. Sacks is asked whether it will be an Ernst & Young-type firm, PwC or KPMG, and his preference is explicit: professional auditors rather than NGOs. That preference is a substantive choice about the character of oversight — accounting-style attestation rather than civil-society review — and it is asserted, not argued.