Dario Defends Himself, Datacenter Panic, AI Doomer Trap, Senate Toss-Up

Dario Amodei did not respond to David Sacks's criticisms for roughly a year and a half, then responded narrowly to a regulatory capture accusation after Gavin Baker's comments appeared on the All In Podcast, without addressing Baker's underlying claims about Anthropic's monopoly ambitions. Sacks argues Anthropic has been more aggressive than any other AI company in promoting public fear, citing three examples: Dario's claim that 50 percent of entry-level knowledge workers would lose their jobs within one to five years, which Anthropic amplified into a media campaign retweeted by President Obama; a study on AI blackmail behavior produced by prompting a model over 200 times until a headline-grabbing result appeared, which the UK AI Safety Institute concluded was engineered under highly pressurized conditions; and Dario and his sister appearing on 60 Minutes to promote that study. The one-year mark on the job-loss prediction has now passed with no recantation, no supporting evidence of AI-driven job loss, and only job creation observed. Sam Altman publicly admitted he was too negative on job displacement while Dario has doubled down.

Sacks draws a sharp distinction between regulatory models. Dario and Demis Hassabis have proposed a FINRA-style body requiring government pre-release approval, which Sacks calls a fig leaf because it is not truly self-regulatory. Dario has also called for FAA and FDA-style frameworks, which Sacks argues would create approval queues, slow US development, and hand China the AI race, citing China's biotech regulatory changes as having already pushed biotech investment offshore and noting China landed a steel rocket while Musk still awaits FAA approval for Starship. The only model Sacks says he could support is the MPAA, which does not report to government and was formed by industry to forestall heavier intervention.

Sacks and others argue an effective open source AI ban is coming but will not be named as such, instead framed as applying the same standards to open models that apply to closed ones. Dario testified around 2023 that sufficiently advanced open models are dangerous because they cannot be centrally monitored, controlled, or rolled back, and the argument is that open models are technologically incapable of complying with centralized safety standards, effectively excluding them. Tomas countered that open source models wrapped in any harness improve in capability while closed models wrapped in their own harness decay, that open source is meaningfully more performant and dramatically cheaper, and that restrictive regulation would cause foreign direct investment to fall sharply and capital to redirect outside America almost instantaneously, drawing a parallel to China's crackdown following Jack Ma's speech before the Ant Financial IPO.

Freeberg explained recursive self-improvement as AI models spinning up agents to collaboratively build a better AI model, eventually becoming fully automated and self-evolutionary. Anthropic has stated it is not certain RSI will happen but believes there is a path to it and is working toward it. Freeberg argued that if RSI becomes viable, regulatory intervention at discrete model development stages becomes a fool's errand because the process becomes continuous, requires only chips, power, and a communication connection, and can run anywhere in the world or in space, including Iceland, Kazakhstan, or ocean platforms outside US jurisdiction. He argued the US would be better off attracting frontier labs under domestic jurisdiction rather than pushing them elsewhere.

Chamath raised concerns about data center growth being constrained simultaneously from multiple directions. Governor Abbott of Texas and Governor Shapiro of Pennsylvania both issued executive orders limiting data center growth. The Treasury Department doubled its bond buying, contributing to rising yields that reduce investor appetite for risk assets including data center investment. Chamath argued frontier model companies are most exposed because they are not investment grade and have weak balance sheets, creating a compounding problem where growth requires compute capacity they cannot fund. Data centers are projected to represent 3 percent of GDP and are credited with generating most of the gains in the stock market, yet public opposition is described as driven more by resentment toward individuals like Jensen Huang and Mark Zuckerberg becoming trillionaires than by the infrastructure itself.

Andreessen Horowitz is under a DOJ investigation according to Bloomberg under Section 8 of the Clayton Antitrust Act of 1914, which prohibits interlocking directorates, involving Ben Horowitz sitting on the Databricks board and a16z partner Martin Casado on the Fivetran board. Speakers noted that opening an investigation file does not mean charges are coming and that the DOJ does not notify subjects when a file is closed, making it effectively a one-way PR weapon. A16z's SpaceX position is estimated at 15 to 17 billion dollars.

Polymarket has Democrats at 52 percent to win the House, making it essentially a toss-up. The current Senate balance is 53 Republican to 47 Democratic seats, requiring Democrats to flip four seats without losing any. Political scientist Patrick Grafini analyzed over 3,000 polls across four election cycles since 2018 and found an average polling error of D plus 3.7, with polls tending to oversample young progressives. Sacks predicted Democrats are favored to win the House but Republicans will likely keep the Senate. A Wall Street Journal article from July 2026 found 53 percent of self-identified conservatives under 40 support government-run grocery stores, and the share of Republicans saying the economic system is rigged toward the wealthy climbed from 30 percent in 2018 to 42 percent, with one speaker predicting a significant movement toward socialist policies by 2028.