Anthropic's Fable Backlash, Nationalizing AI, Inflation Heats Up & California's Broken Elections
The All-In podcast hosts discuss Anthropic's Claude 'Fable 5' model release and its controversial surveillance and content-downgrading policies, Bernie Sanders' proposal to seize 50% equity in AI companies for a sovereign wealth fund, hot inflation prints driven partly by the Iran war, and alleged electoral fraud in the Los Angeles mayoral primary involving ballot harvesting.
Summary
The episode opens with a deep dive into Anthropic's release of 'Fable 5,' described as a top-benchmarking model that costs twice as much per token as its predecessor. The hosts focus on two major controversies: first, Anthropic retains all prompt data for at least 30 days with no exceptions, even for enterprise customers with zero-data-retention agreements; second, the model secretly downgrades users it detects as doing frontier AI research, serving them a lesser model without disclosure while still charging full price. Anthropic has since walked back the non-disclosure aspect but still maintains the downgrading practice. The hosts argue this represents a dangerous precedent of AI censorship and surveillance, with Sacks framing it as regulatory capture — Anthropic simultaneously restricting competitors via model policies while lobbying for government regulation that would effectively ban open-source models. Freeberg shares that his genomics company, Ohalo, has already been impacted, as restrictions on biological research prompts are forcing them toward open-source Chinese models like DeepSeek, which the hosts argue is a self-defeating outcome for American competitiveness. Chamath warns of a corporate favoritism risk where AI companies could subtly shape outputs to benefit strategic partners. The hosts steel-man Dario Antropic's position as a genuine believer in AI risk, noting that this worldview attracts top AI talent, but ultimately conclude the hypocrisy — building recursive self-improvement while calling for a pause — undermines credibility.
The second major topic is Bernie Sanders' op-ed proposing the 'American AI Sovereign Wealth Fund Act,' which would impose a one-time 50% tax on the stock (not profits) of major AI companies, with shares placed into a government fund giving citizens voting rights and board representation. Sacks notes that while he opposes confiscation, the political logic is understandable given AI CEOs repeatedly telling the public that AI will destroy jobs. Freeberg argues there is no significant job loss from AI — his own company is actively hiring more because AI expands revenue opportunity rather than just cutting costs — and proposes instead reforming Social Security into an equity-based sovereign wealth fund that invests in American companies including AI firms. Chamath points out AI economics differ fundamentally from the internet: every marginal AI user has real compute costs, unlike zero-marginal-cost internet users, giving government a legitimate infrastructure argument for ownership. The hosts note irony that OpenAI and Anthropic are public benefit corporations that have invited this political scrutiny through their own doom-and-gloom messaging.
The third topic covers May's CPI print of 4.2% year-over-year (highest since April 2023) and PPI at 6.5% (highest since late 2022), with Polymarket pricing a 49% chance of Fed rate hikes this year. The hosts attribute much of the spike to energy price pressure from the Iran war and link it to uncontrolled government spending. Chamath notes China's energy reserves have helped suppress oil prices but warns of a spike risk if China returns to spot markets. Sacks notes markets were actually up on the day, suggesting expectations of an Iran resolution.
The final segment covers the Los Angeles mayoral primary, where Spencer Pratt led on Election Day in-person votes (35%) but fell behind Nithya Raman in post-election mail-in ballots. The hosts present statistics showing Pratt's share of late mail-in ballots dropped by one-third while Raman's surged 80%, which they argue is statistically near-impossible organically. Freeberg outlines how California's layered laws — universal ballot mailing, unlimited ballot harvesting, no voter ID, minimal signature verification, and new restrictions on audits signed by Newsom — create a legal framework that enables appointment rather than genuine election. Sacks calls it 'legal fraud.' The hosts call on the Trump DOJ to investigate and support a pending California voter ID ballot measure as a last democratic remedy.
Key Insights
- Freeberg argues that Anthropic's content restrictions on biological research are already forcing his genomics company to adopt open-source Chinese AI models, which he says damages American competitiveness in a self-defeating way.
- Sacks contends that Anthropic is engaged in regulatory capture — simultaneously restricting competitor capabilities through model policies while lobbying for government AI regulation designed to effectively ban open-source models.
- Chamath warns that beyond censorship risk, enterprises face a 'rug pull' business risk from AI providers who can arbitrarily cut off access to capabilities they depend on, making single-provider AI dependency a strategic vulnerability.
- Freeberg asserts that AI is creating far more jobs than it eliminates by enabling companies to expand their revenue and product surface area, pointing to strong May jobs numbers as empirical evidence against AI doomsday narratives.
- Sacks argues that Dario Altman's repeated public statements about massive AI-driven job loss have politically justified Bernie Sanders' proposal to seize 50% of AI company equity, meaning the AI companies effectively invited this political backlash.
- Chamath argues AI economics are fundamentally different from the internet because every marginal AI user has real compute costs, unlike zero-marginal-cost internet users, giving government a legitimate infrastructure ownership argument.
- Freeberg outlines how California's stacked election laws — universal ballot mailing, unlimited ballot harvesting, no voter ID, weak signature verification, and new audit restrictions — collectively create a legal system of appointment rather than free democratic election.
- The hosts present statistics showing Spencer Pratt's share of post-election mail-in ballots dropped by one-third while Nithya Raman's surged 80%, which they argue is statistically near-impossible without organized ballot harvesting operations.
- Sacks notes that Anthropic secretly downgraded users to lesser models without disclosure while still charging full price, and only walked back the non-disclosure aspect after public outcry — not the downgrading practice itself.
- Chamath argues that a gigawatt of AI compute capacity now costs approximately $100 billion, up 20x from his initial estimates, creating a massive capital moat that makes meaningful open-source compute infrastructure practically inaccessible to private actors.
- Freeberg proposes reforming Social Security from a defined-benefit treasury-only system into an equity-based sovereign wealth fund that invests in American companies including AI firms, giving every citizen account-based ownership stakes.
- Sacks draws a parallel to early 2020s social media content moderation, warning that AI companies' expansive definitions of 'safety' — already flagging mitochondria questions and cancer/GLP-1 research — follow the same pattern of mission creep that led to political censorship on social platforms.
Topics
Full transcript available for MurmurCast members
Sign Up to Access