Good morning. It's Friday, June 12, and we're covering China's bid to dominate humanoid robotics, Anthropic's move toward a tiered AI future, and a proposal to give the public a stake in the AI economy.

Plus: yesterday's Musk poll results, and a guest essay on what happens when ChatGPT, Nancy Pelosi, and curiosity become your investing guides.

Anthropic is embracing an unusual idea: giving Americans a financial stake in the AI boom rather than limiting gains to investors. CEO Dario Amodei's new policy framework proposes a public fund that could let citizens share in the value created by leading AI companies. The plan arrives as Washington debates government equity stakes in AI firms and broader responses to AI-driven job disruption.

OpenAI is preparing for an AI pricing battle by considering steep token cost cuts. The company is discussing lower rates for the usage-based units that power its models to strengthen its position against Anthropic. The plans remain unsettled and are partly driven by expectations that Anthropic will also reduce prices, according to people familiar with the talks.

Anthropic is backing away from a little-known Claude Fable 5 safeguard after criticism over its lack of transparency. The company says it will make protections affecting frontier AI research visible instead of quietly degrading performance for some users. Anthropic acknowledged it "made the wrong tradeoff" and apologized following backlash over the policy. The reversal highlights growing scrutiny of how AI companies enforce model restrictions.

SpaceX priced 555.6M shares at $135 each, raising $75B in the largest IPO in history (3x Saudi Aramco's 2019 record), with trading set to begin Friday on Nasdaq as SPCX. The book was four times oversubscribed, and synthetic exposure on Hyperliquid is already pricing SPCX at $167 (a classic ~20% pop), giving underwriters the option to release another 83.3M shares for ~$11B more. Musk becomes the world's likely first trillionaire on the strength of ~850M Class A shares plus 5.6B Class B shares (10 votes each, 1B contingent on a million-person Mars colony).

FRIDAY FACTS

Almost every website you visit asks for your permission for something invented in 1994 by a 23-year-old engineer trying to solve a single technical problem for a single client. The story is at the bottom of today's newsletter.

AI's $1.3M token bill wasn't for coding—it powered a "software factory" that automated thousands of GitHub issues and PRs.

Matthew Berman recently asked me how much I thought a prominent AI engineer had been hired for. I said a billion dollars. A few weeks later, he asked me what it cost to run GTC (one of NVIDIA’s tech conferences). I said a billion. He asked what SpaceX might IPO for. I said a billion again. In my head, a complex financial calculus was occurring: how much money should I have before retirement? A billion. (Cue Dr. Evil meme.) → Read the full article here.

SemiAnalysis argues that China's Unitree is following a BYD-style playbook to become the dominant force in humanoid robotics through aggressive cost reductions, vertical integration, and rapid hardware iteration. The report says Unitree could ship its 10,000th humanoid robot within weeks, has tripled revenue year over year, plans to spend nearly $300 million on AI research and development, and has cut its flagship G1's price from more than $50,000 to about $27,300 while maintaining strong margins.

The authors contend that improving reliability and lower costs are pushing the G1 beyond research and entertainment into real-world labor deployments, with more than 250 units already at work in commercial settings by their estimates. Drawing parallels to BYD's rise in electric vehicles, the report suggests Unitree's control over key components and supply chain could help create new robotics markets and strengthen China's manufacturing advantage over Western competitors. → Read the full article here.

Anthropic's launch of Claude Fable 5 has shifted attention from benchmark gains to a new access model that separates public and restricted frontier AI capabilities. Fable 5 is the publicly available version of the company's Mythos-class system, but high-risk requests involving cybersecurity, biology, chemistry, and AI distillation can be routed to the older Claude Opus 4.8, while the less restricted Mythos 5 is reserved for vetted organizations through Anthropic's trust program.

Anthropic says the approach balances safety with broader access, but the rollout highlights a growing divide between general users and approved institutions with access to more capable systems. The launch also underscores the economics of frontier AI: Fable 5 is included with paid Claude plans through June 22, after which it will temporarily require usage credits until Anthropic has sufficient computing capacity to restore broader access. → Read the full article here.

Waymo is introducing Premier, a subscription program that offers frequent riders perks including priority pickups, 10% Waymo Cash back on trips, free monthly cancellations, and early access to service in new cities. The membership is designed to strengthen Waymo's direct relationship with customers as the company expands into markets such as Dallas, Houston, Miami, and Orlando through its own app rather than ride-hailing partners.

A Waymo spokesperson said the initial number of Premier members will be limited to minimize the impact on nonmembers' wait times. The move comes as Waymo and Uber pursue different strategies for autonomous transportation, with Waymo building a standalone robotaxi ecosystem while Uber continues to argue that a hybrid network of human drivers and autonomous vehicles will be more scalable. → Read the full article here.

India's Filmmakers Lean Into AI: JioStar and other studios are using generative AI to cut costs and speed production as demand for content surges.

Amazon Locks In $17.5B AI Loan: Amazon secured flexible financing to fund AI expansion, joining Big Tech's shift toward debt-backed infrastructure.

Meta Cuts Ties With Manus (Paywall): Meta is shutting down the AI platform and separating its systems after Beijing ordered the $2 billion deal unwound.

Anthropic Funds AI Fellowship: Anthropic is putting $150 million behind an AI workforce experiment that pays fellows to work in nonprofits for a year.

Deezer Launches AI Music Detector: Deezer's free tool scans playlists for AI-generated songs as streaming platforms face fraud and copyright concerns.

BlackRock Eyes $5B in SpaceX IPO: BlackRock reportedly sought a major stake as SpaceX targets a record $75 billion IPO at a $1.8 trillion valuation.

Here’s how you voted:

🟨⬜️⬜️⬜️⬜️⬜️ Genius — make your rivals pay your company's bills (8%)
🟩🟩🟩🟩🟩🟩 A pump — circular financing dressed as revenue (47%)
🟨🟨🟨🟨⬜️⬜️ Both — and SpaceX wins either way (33%)
🟨⬜️⬜️⬜️⬜️⬜️ This ends badly for someone (11%)

FRIDAY FACTS

In June 1994, a 23-year-old engineer at Netscape named Lou Montulli was handed a problem by a client, MCI. MCI wanted to build an early online store, but it didn't want its servers to remember what was in any one shopper's cart between page loads — that approach wouldn't scale. Montulli, working with colleague John Giannandrea, wrote a spec for a small piece of data the server could store on the user's computer instead. He called it a cookie, borrowing from the Unix term "magic cookie." Netscape shipped support for it that October.

Every cookie consent banner, every "remember me" checkbox, every ad that follows you around the internet traces back to that fix. Thirty years later, Montulli's quick solution is the substrate on which the entire stateful, personalized web is built — and the reason you click "Accept All" a dozen times a week without thinking about it.

Thanks for reading. See you next time!

— Matthew Berman, Nick Wentz & the Forward Future Team