All-In PodcastAnthropic's Fable Backlash, Nationalizing AI, Inflation Heats Up & California’s Broken Elections
CHAPTERS
Besties reunite and set the agenda: Anthropic, regulation, macro, and elections
The hosts return as the full quartet and tee up a packed episode focused on AI platform power, political responses to AI, the macro inflation picture, and California election integrity controversies. Jason frames the episode around escalating stakes: model capability is rising while trust and governance are eroding.
Anthropic’s “Fable 5” launch sparks backlash: prompt retention, privacy, and secret downgrades
Jason summarizes the release of Anthropic’s new top-benchmark model “Fable 5,” including higher token costs and strict safety guardrails. The controversy centers on mandatory 30-day prompt/output retention and the discovery that certain users were silently downgraded if Anthropic suspected frontier-model development or sensitive research.
Enterprise risk and “AI censorship”: why companies fear rug-pulls by closed labs
Chamath argues Anthropic’s approach reveals a new operating model: pre-screen prompts, classify users, and decide what capabilities to provide—raising censorship and business continuity risks. For enterprises, accidental triggering of safeguards can become a single point of failure for core workflows, forcing a shift toward diversified providers and governance planning.
Friedberg’s biotech use case: safety filters block legitimate science and push teams to open source
Friedberg describes using LLMs for genomics workflows (construct design, guide RNA design, phenotype prediction) and how recent restrictions have degraded that utility. He argues censorship will push serious teams to run models locally, start building their own models, and—critically—adopt the best available open-source options, which he claims are often Chinese.
Regulatory capture thesis: surveillance + nerfing + calls for an AI “FAA/FDA”
Sacks reiterates his view that Anthropic is pursuing regulatory capture by fearmongering while implementing surveillance and access tiering. He highlights user outrage over mandatory data retention and undisclosed downgrades, warning that if the same firms lobby for licensing regimes, competition and open-source alternatives could be throttled.
Pragmatic safety: regulate downstream harms (e.g., gene synthesis screening), not model access
The group debates steelmanning Anthropic’s safety posture while arguing for more targeted interventions. Sacks and Friedberg point to nucleic-acid synthesis screening and recordkeeping as a workable safeguard: intervene when model outputs are translated into physical bio-risk actions, rather than broadly restricting general research queries.
Compute, power, and the capital moat: open source needs megawatts to compete
Chamath shifts the discussion to infrastructure: even if open source exists, the bulk of compute is directed to big closed models. He argues sustaining a vibrant open-source ecosystem requires large-scale power and data center buildout, but the cost of a gigawatt-scale facility has ballooned, creating a massive barrier to entry.
Nationalizing AI / sovereign wealth fund proposals: Sanders, Trump, and “who should own AI?”
Jason introduces Bernie Sanders’ proposal to take 50% equity in major AI firms into a public sovereign wealth fund with voting rights and board representation. The hosts reject outright confiscation but acknowledge political logic: AI trained on public data and predicted to disrupt jobs invites demands for public ownership or benefit-sharing.
Reforming Social Security into an investment sovereign fund (and disputing the ‘AI job apocalypse’)
Friedberg argues the best mechanism for broad benefit-sharing is structural: reform Social Security to hold equities and distribute ownership via accounts rather than promises. He strongly disputes imminent mass job losses, claiming AI primarily expands revenue/productivity, increasing hiring needs rather than driving a near-term employment collapse.
Liquidity conference recap: standout speakers, venture data, and event positioning
The hosts recap highlights from Liquidity, including praise for OpenAI CFO Sarah Friar and investor Thomas Lefont’s data-driven VC presentation. They position Liquidity as a curated allocator forum versus the broader Summit as a festival-like event, and briefly acknowledge sponsors and programming plans.
Inflation heats up: CPI/PPI prints, energy shocks, and rate hike odds
The hosts react to hot CPI and PPI readings, connecting the spike partly to energy shocks tied to geopolitical conflict and partly to long-run fiscal overspending. They discuss the market’s muted reaction (stocks up), possible Fed rate paths, and how oil pricing—especially China’s buying behavior—could amplify inflation.
California election integrity controversy: mail-in dynamics, ballot harvesting, and trust collapse
The group debates Los Angeles election results and the statistical patterns between in-person voting and late-arriving mail ballots. Friedberg and Sacks argue California’s election rules (mass mail ballots, loose verification, ballot harvesting) create a system vulnerable to manipulation—legal or illegal—eroding faith in democratic legitimacy.
