Jun 13, 2026, 1:10 AM ETInterview, Podcast

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

Anthropic released "Fable 5" (a "mythos-level" model) on Tuesday, which topped nearly every benchmark test.
Token costs for Fable 5 are twice as high as the previous Opus 4.8 model, though Anthropic claims overall token usage will be lower due to superior efficiency.
The model includes strict guardrails blocking topics like bio-weapons and hacking, a response to early access programs where the tool was used to seal vulnerabilities at Palo Alto Networks.
A significant portion of the backlash stems from a privacy clause where prompt data is retained for at least 30 days, applicable even to enterprise customers with zero-data agreements.
Anthropic initially downgraded users engaged in frontier AI research (e.g., making competitive models) without disclosure; this practice was hidden in a 319-page document.
Following public outcry, Anthropic stated it would change safeguards to make downgrades for frontier LLM development "more visible," though it continues to restrict access.
Specific instances of arbitrary downgrading were reported, including a user asking about mitochondria and Ben Thompson (Stratechery) asking about cancer risks and GLP-1 drugs.
Critics argue these actions create an "AI have and have-not" system where Anthropic profiles users to decide what capabilities they deserve, potentially acting anti-competitively.

David Freiberg (Recursion Pharmaceuticals) warns that Anthropic's restrictions on genomics and gene-editing tools are forcing companies to abandon closed models in favor of local, open-source alternatives.
Freiberg notes that the best current open-source models are Chinese, creating a geopolitical risk where US regulatory pressure inadvertently advantages China in biotech and material science.
Freiberg predicts a shift where companies will train and run their own proprietary models locally to avoid "rug pulls" or censorship from major API providers.
Chamath Palihapitiya identifies the core risk as Anthropic taking the "holy grail" of AI safety by evaluating prompts before generating output, creating a single point of failure and censorship risk for enterprises.
Sax Sethi argues that Anthropic is engaging in "regulatory capture" by fear-mongering about AI risks to justify restrictions that protect their market share while calling for a new regulatory agency (like an FAA) to approve models.
The discussion highlights a "preemptive strike" against open-source models, as regulatory frameworks proposed by Anthropic's leadership would be impossible for decentralized open-source entities to satisfy, effectively killing competition.
Investors are being urged to consider building large-scale compute infrastructure (e.g., 2 gigawatt data centers) to support a sovereign open-source model ecosystem, though the capital cost (~$100 billion per gigawatt) is prohibitive.

Senator Bernie Sanders proposed the "American Sovereign Wealth Fund Act," calling for a one-time 50% tax on equity of major AI companies (OpenAI, Anthropic, xAI) to create a public trust.
Sanders argues that AI is built on "stolen" collective human intelligence (public data) and that the public should own half the resulting wealth.
Chamath Palihapitiya rejects the "confiscation" aspect but agrees with Freiberg's counter-proposal: reforming the Social Security Trust Fund to invest directly in equity (including AI companies) rather than holding only Treasury bonds.
Freiberg contends that a sovereign wealth fund should be an investment vehicle where citizens own shares via a defined-contribution system, rather than a mechanism for asset seizure.
Sethi argues that AI CEOs are asking for this regulation by claiming they will cause 50% job losses; since they train on public data and gatekeep access, public ownership is a logical political reaction.
Chamath notes the economic difference between the internet (near-zero marginal cost) and AI (high marginal GPU/energy costs), which complicates the "public good" argument but suggests the government has leverage similar to owning a percentage of interstate highway users.

May CPI data came in at 4.2% year-over-year (highest since April 2023), while PPI rose to 6.5% year-over-year (highest since late 2022).
Chamath Palihapitiya attributes the inflation spike to "excess government spending" driven by political incentives to give citizens more than they currently have.
Freiberg cites a potential energy blip from the Iran war as a factor but warns that if China exhausts reserves and re-enters the spot market, oil prices could spike to $150–$200, causing severe downstream inflation.
Polymarket odds currently show a 49% chance of a Fed rate hike later this year, up from under 10% prior to the Iran escalation.
The hosts anticipate a "late stage" of higher rates, with overnight rates potentially reaching 5.5%–6%, as fiscal deficits drive monetary tightening.

Freiberg presents statistical anomalies in the LA Mayoral primary showing Nithya Raman gained 80% of her late-arriving mail-in ballots compared to a 33% decline for Spencer Pratt's late ballots.
Freiberg argues the system is "designed for fraud," citing California laws on unlimited ballot harvesting, lack of voter ID, and the mailing of ballots to non-existent addresses.
Sethi counters that while individual acts of fraud (e.g., paying for votes) are illegal, the specific laws enabling ballot harvesting and loose registration are legal, creating a "design of appointment" rather than a true election.
Freiberg suggests the "cheating operation" for Raman was more sophisticated due to existing ballot harvesting networks, contrasting with Pratt's lack of a ground game.
Chamath and Sethi acknowledge loopholes exist (e.g., gym memberships for registration, no ID for voting) but argue that major election fraud swinging results is statistically rare and usually results in lawsuits rather than systemic collapse.
Freiberg concludes that the media is silencing discussion to avoid being labeled "Trump-pro," preventing an honest statistical debate about the election integrity.
Freiberg proposes that a declaration of "State of Emergency" by a top-level official (specifically naming Steve Hilton) is the only "break glass" way to bypass the current election laws and clean up the system.