All-In Podcast

Anthropic's Fable Backlash, Nationalizing AI, Inflation Heats Up & California’s Broken Elections

1h 42mJun 13, 2026

Key Themes

Frontier AI governanceOpen-source accessPublic ownership debatesInflation pressuresElection integrityInstitutional trustAI infrastructure constraints

Summary

The hosts connect AI safety backlash, public-control proposals, inflation anxiety, and California election disputes into a broader argument about trust in institutions.

This episode centers on whether powerful institutions—frontier AI labs, regulators, central banks, governments, and election systems—are earning public trust. The hosts begin with the backlash to Anthropic’s Fable 5 release, criticizing prompt retention, selective model downgrades, and safety rules they fear could chill research or push users toward open-source alternatives. They then debate whether AI safety concerns should be handled through broad model-access restrictions or downstream safeguards, before turning to Bernie Sanders’ proposal to take equity in major AI companies for a sovereign wealth fund. Later chapters cover the All-In Liquidity event, hotter inflation readings and energy-price risks, and a heated discussion about California election rules, mail-in ballots, ballot harvesting, voter ID, and doubts around an LA primary vote-count pattern.

Trust is the common thread across very different controversies

Whether discussing AI model restrictions, sovereign wealth fund proposals, inflation, or election rules, the hosts keep returning to the same question: who gets to control powerful systems, and what safeguards make those systems legitimate to the public? The episode’s debates are less about one policy than about institutional credibility.

+1 more

AI safety debates are moving from technical design to power and governance

The Anthropic discussion shows how AI safety choices—prompt retention, model downgrades, access controls, and monitoring—can be interpreted not only as risk mitigation but also as censorship, surveillance, or market control. The hosts argue that the placement of safeguards matters: broad access restrictions create very different social consequences than downstream checks on dangerous real-world outputs.

+1 more

Open-source AI is presented as both a safety challenge and a freedom valve

The hosts describe open-source models as difficult to regulate once widely distributed, but also as an important alternative when closed models retain data, degrade answers, or block sensitive research. This tension captures the broader dilemma: openness can increase misuse risk, yet it can also preserve competition, research access, and user autonomy.

+1 more

Calls to nationalize AI reflect political backlash to private control of public knowledge

The hosts reject Bernie Sanders’ proposed 50% stock tax as confiscatory, but they acknowledge the political logic behind it: AI labs use public knowledge, public infrastructure, and scarce energy and chip resources while some leaders warn that the technology may displace workers. The debate highlights how AI’s social bargain remains unsettled.

+1 more

The hosts push back on simple AI job-loss narratives

Instead of treating AI as only a labor-destroying force, the hosts argue it may expand productivity, create new revenue opportunities, and change what companies can attempt. That distinction matters because public fear of job loss is part of what fuels demands for regulation, taxation, or public ownership.

Inflation concerns remain tied to fiscal policy, energy shocks, and expectations

The inflation segment frames hot CPI and PPI readings as part of a broader concern about government spending, potential Fed action, and energy-price spillovers from geopolitical conflict. Even when markets do not immediately panic, the hosts treat inflation as a structural issue that can reappear through energy and input costs.

+1 more

Election confidence depends on both legality and perceived legitimacy

In the California election debate, the hosts distinguish between illegal fraud and legal practices that can still undermine trust, such as broad ballot collection, weak verification, and long post-election counting windows. Jason’s request for a steelman explanation underscores that confidence requires both rigorous procedures and serious scrutiny of unusual patterns.

+1 more

Select any chapter text to Deep Dive with AI

01Anthropic Fable Backlash: Privacy, Model Nerfing, and the Push Toward Open Source AI

The hosts open by debating backlash to Anthropic’s Fable 5 model, focusing on prompt retention, undisclosed downgrades for some research uses, privacy concerns, and the possibility that safety rules can become censorship or anti-competitive gatekeeping. They argue that these choices may push enterprises, biotech researchers, and developers toward local or open-source models, including Chinese open-source alternatives, while also acknowledging the real risks of AI-enabled cyber, biological, or physical harm.

Anthropic’s Fable 5 is described as a high-performing model whose release drew criticism over 30-day prompt retention and capability restrictions for frontier research use cases.

The hosts frame prompt evaluation and selective model access as privacy, enterprise-governance, censorship, and single-point-of-failure risks.

Friedberg says restrictions on biological and genomic work are already pushing companies toward locally run open-source models.

The group worries that AI regulation could entrench closed incumbents by burdening or limiting open-source competitors.

Chamath emphasizes the enormous infrastructure cost of open-source AI at frontier scale, citing gigawatt data-center costs.

The discussion acknowledges dual-use risks but argues that safeguards should target harmful real-world outputs rather than broadly restricting access to AI tools.

02AI Regulatory Capture vs. Pragmatic Safety Guardrails

The second chapter sharpens the regulation debate: the hosts warn that safety rhetoric could become a pathway to regulatory capture, leaving a few companies and a government agency in control of advanced AI access. They contrast broad user profiling, access controls, and model downgrades with targeted downstream safeguards such as screening DNA or RNA synthesis orders for bioweapon risks.

The hosts argue that AI safety regulation could become a monopoly or duopoly protection mechanism.

They criticize KYC-style gating, user profiling, and model downgrades as potential surveillance or censorship tools.

Nucleic-acid synthesis screening is presented as an example of a practical downstream safeguard against misuse.

Open-source models are compared to published knowledge: once widely distributed, they cannot realistically be recalled or turned off.

The hosts suggest users will migrate away from AI platforms that block or degrade ordinary queries, unless alternatives are also restricted.

03Nationalizing AI and Sovereign Wealth Fund Proposals

The hosts debate Bernie Sanders’ proposal for a one-time 50% stock tax on major AI companies to create an American AI sovereign wealth fund. They largely oppose equity confiscation, but they also explore why the politics of public ownership may become more attractive if AI leaders warn about mass job loss while training on public human knowledge and relying on public infrastructure, power, chips, and data.

Sanders’ proposal is described as taking 50% of stock in large AI companies such as OpenAI, Anthropic, and xAI for a sovereign wealth fund.

Sacks rejects the idea as property confiscation while arguing AI leaders have invited backlash by warning of widespread job losses.

Friedberg proposes reforming Social Security into account-based public investment rather than seizing company equity.

The hosts dispute extreme AI job-loss narratives, arguing AI may expand productivity and new opportunities.

Chamath argues AI differs from internet businesses because each marginal user consumes costly GPUs, electricity, and memory.

The group suggests open-source obligations may be a better remedy than equity seizure when closed AI labs train on public knowledge but restrict outputs.

04Liquidity Recap and Hot Inflation Print

The hosts recap their Liquidity event, praising OpenAI CFO Sarah Friar and highlighting Thomas Laffont’s data on venture outcomes. They then turn to hotter-than-expected inflation readings, discussing CPI, PPI, Fed hike odds, energy shocks linked to the Iran war, government spending, and why markets appeared to treat the inflation print as broadly expected rather than catastrophic.

Sarah Friar’s Liquidity interview is praised as one of the standout moments from the event.

Thomas Laffont’s data presentation is summarized as showing stronger scaling odds for later-stage mega-winners than for smaller unicorns.

Liquidity is described as a curated gathering for major capital allocators, distinct from the broader All-In Summit.

The hosts cite hot May CPI and PPI readings and discuss higher odds of a Fed hike.

Friedberg ties persistent inflation and inequality pressures to excess government spending.

Chamath warns that energy-market disruptions could push oil prices sharply higher and feed into CPI through global input costs.

Sacks notes that markets were up despite the data, suggesting the print was not treated as a severe surprise that day.

05California Election Rules, Ballot Harvesting, and LA Primary Integrity Doubts

The final chapter is a contentious debate about California election procedures and an LA mayoral primary vote-count pattern. The hosts discuss whether late-arriving mail-in ballots, ballot harvesting, weak verification, and voter-roll issues reflect illegal fraud, legal but confidence-eroding tactics, or a sophisticated campaign ground game. They call for audits, stronger verification, voter ID reforms, and more serious media scrutiny, while Jason presses for a legitimate steelman explanation of the discrepancy.

The hosts cite a sharp difference between in-person, earlier mail-in, and later-arriving mail-in ballot patterns in an LA primary.

Friedberg argues the issue may be less illegal fraud than a legal structure that enables broad ballot collection and weak confidence in outcomes.

Sacks criticizes universal mailed ballots, voter-roll quality, signature verification, voter ID rules, and chain-of-custody gaps.

Chamath frames California as a one-party political machine that has legalized practices he sees as improper.

Jason pushes back by asking whether there is proven election-changing fraud and argues credible claims should lead to audits or investigations.

The hosts broadly discuss reforms including voter ID, ending universal mailed ballots, and stronger verification requirements.

A steelman explanation offered is that one campaign may simply have had a better ground game and ballot-collection operation than a first-time candidate.