BRIEF
Anthropic's leaked IPO papers show $518 billion in spending commitments, and 80% of it can't be canceled.
Jacob Coxon, a former Anthropic staffer, went public on September 9 with warnings that AI could destroy humanity. Two weeks later the Department of War declared the U.S. "Will Never Be an Effective Altruist Country."
Anthropic's preliminary Q2 revenue of $11.6 billion softened the math.
It did nothing for the bigger problem: the people grading Anthropic's homework share the belief system that wrote it.
Who are the believers?
Photo illustration by Shirley Yu at Sum of Parts for Zero Knowledge
Good morning,
Markets can't call interest rates ten years out.
But Effective Altruists price AI outcomes fifty years out.
Straight-faced. No error bars.
Now they've committed $518 billion on the strength of it.
Grumping,
Austin Campbell
Teaching DeFi at NYU Stern
Bridging TradFi at Zero Knowledge
P.S. The Zero Knowledge team works hard to produce this for you.
If you like it, share it with your colleagues.
🏹 BACK UP
WTF IS EFFECTIVE ALTRUISM?
Effective Altruism is a utilitarian movement that tries to do the most measurable good with money and effort, and it shaped the founding culture of both Anthropic and OpenAI.
“Is Effective Altruism Now Defective Thanks to SBF?” New York Magazine Nov 2022 (Featuring Sam Bankman Fried
$500M
WTF IS OPEN PHILANTHROPY?
Open Philanthropy is the grant-maker, funded by Facebook cofounder Dustin Moskovitz and his wife Cari Tuna, that became the biggest single funder of Effective Altruist causes. It rebranded as Coefficient Giving in 2025.
Dustin Moskovitz and Cari Tuna
62.5%
WHO IS JACOB COXON?
Jacob Coxon is the former Anthropic researcher who quit and went public on September 9 with warnings that out-of-control AI could destroy the human race. He says there is a 10% chance AI "could kill us all by the end of the decade."
The Anonymous Math Geek Who Quit Anthropic—and Became the Face of AI Safety - WSJ September 2026
2 WEEKS
⏳ BREAKDOWN
In under three weeks, a resignation turned into an audit of Anthropic's books.
SEP 9
Coxon quits and goes public with warnings that AI could destroy the human race, in a Wall Street Journal profile. He later said he "misspoke" when he told Fox's Bret Baier he hadn't coordinated the announcement.
SEP 28
Reuters reports it obtained a copy of Anthropic's IPO prospectus. Losses grew from $2.98B in 2024 to $8.06B in 2025.
SEP 29
Reuters details $518B in future spending and lease commitments. About 80% can't be canceled, so they get paid whether or not Anthropic ends up needing the data centers. The filing also warns the models could pose "existential risks to humanity."
SEPT 30
Anthropic reports preliminary Q2 2026 revenue of $11.6B, against $4.6B in gross revenue for all of 2025. Unaudited, so a grain of salt applies.
OCT 4
OpenAI CEO Sam Altman tells Politico: "We believe the world should accept some bad things happening for the benefits of this technology." [NEW]
NEXT
Anthropic's IPO is reportedly planned for November. OpenAI has said its own listing slips to 2027.
ZERO INSIGHTS : THE LONG READ
ALTRUISM, INC.
$518 billion in commitments, built on a belief system. David Z. Morris on the Effective Altruist worldview behind Anthropic's IPO, and why the people grading the homework wrote it.
🎭 BUSTED
THEY SAY
AI safety is a science, and what they do is research.
BUT ACTUALLY
A 2019 Open Philanthropy grant to the Effective Altruism Foundation came with "Communication Guidelines" that steered researchers away from saying the future could be bad overall, according to two whistleblowers. One grantee, Simon Knutsson, says he was told he could likely get future funding if he followed them. He declined.
If your funder has to remind you they’re not asking you to lie, something has already gone very wrong.
💰 BAG HOLDERS
WINNERS
Alphabet, Amazon, Microsoft, Broadcom, SpaceX, and AMD, the counterparties on the $518B in commitments, who get paid if Anthropic can.
Open-weight model builders, whose models cost a fraction of Anthropic's.
Effective Altruism's critics, left and right, who suddenly share a megaphone with the Department of War.
LOSERS
IPO buyers, who'd own the equity sitting behind $518B of mostly non-cancelable commitments.
Towns hosting data centers that may never be fully used.
The public, if David's worry holds and AI-lab debt turns into a bailout question.
Dissenting researchers, like Simon Knutsson, who turned down the guidelines and now takes carpentry and warehouse work to make ends meet.
ZERO OUT
Anthropic's numbers got better in a week but who verifies them didn't.
BEST CASE
The communication-control habits are a relic, the labs grow up, and hacks, breaches, and security failures get reported straight.
WORST CASE
The labs keep grading their own homework, the IPO prices on belief, and the commitments outrun the revenue.
1️⃣ Watch the public S-1 ahead of the reported November IPO: do the $518B and the 80% non-cancelable figure survive audited numbers?
2️⃣ Test revenue against the commitments each quarter. The $11.6B in preliminary Q2 revenue is the number to check, and it's unaudited.
3️⃣ Watch who discloses the next breach or security failure at a lab: the lab itself, or someone independent.
ZERO INSIDER
One of my favorite conversations of all time : Our ZK studios hosted Adam Zarazinski, CEO of the intelligence firm Inca Digital for Zero In's Episode 1 : Scams - hilarious, educational, and on executive time (~5-7 minutes).
David appeared on the TrueAnon podcast to discuss the "Effective Altruist" movement and artificial intelligence.
& David joined Ryan Research podcast for a finance discussion of the same.