The AI Doom Narrative Is a Psyop, and the Target Is Open Source
Anthropic asks public markets to underwrite it at a trillion-dollar valuation while its safety lead endorses the claim that AI may end civilization — not an internal contradiction, but regulatory groundwork aimed ultimately at open source.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
The resignation letter wasn't spontaneous — it was organized PR
Jacob Coxon quit Anthropic after only six weeks and posted that the people building AI genuinely believe it could kill everyone by the end of this decade. Sachs points out this was not spontaneous: the account had almost zero followers and zero posts before, yet got hundreds of millions of views in a day; within the first 15 minutes three Doomer organizations amplified it, all funded by the same big EA donor, who also co-led Anthropic's Series A. More tellingly, the Wall Street Journal story went out minutes before the tweetstorm — meaning the reporter had been briefed under embargo. Sachs says if Coxon really came on the show, the first question would be: who introduced you to the Wall Street Journal?
— David SachsThe company didn't distance itself — it amplified him
A normal company facing a whistleblower immediately pushes back and distances itself. Anthropic did the opposite: Evan Hubinger, who runs alignment, publicly replied that Jacob is right, that they genuinely believe AI could kill all of humanity, and that he personally puts the probability above 10% over the next decade. Sachs argues this exposes the contradiction — what he said matches the company's internal Doomer agenda exactly, which is why he got internal support. Chamath adds the knife: the company can't distance itself, because the moment it does, that whole internal bloc blows up.
— David Sachs / ChamathAsking for a trillion-dollar valuation while saying the product could end civilization
Sachs lays out Anthropic's IPO dilemma clearly: you are asking public-market investors to underwrite your company at a trillion-dollar valuation while your safety lead says the core product is unsolved and could end civilization. That is at minimum the seedbed of product-liability litigation. Chamath uses the Philip Morris analogy: we know cigarettes kill you, we'll say so, but we won't disclose it, we'll still go public and let you buy the stock, and we'll figure it out later. Sachs calls it "a time for choosing" — either publicly deny that Coxon is a Doomer operation, or admit Bernie Sanders is right and development should be frozen, in which case how do you IPO at a trillion-dollar valuation?
— David Sachs / ChamathThe real target of regulation is open source
Sachs is blunt: this is a Doomer psyop, and the end goal is open source. Because FDA-style regulation won't say "ban open source" — it will say "the same standards apply to open and closed models" — and Dario has already testified in the Senate that models are dangerous if they can't be centrally monitored and rolled back. Once open-model weights are released they can't be rolled back, so the standard landing is equivalent to banning the release, use and hosting of those weights. Freeberg adds: open source has no corporate structure to file regulatory paperwork, and open source is precisely what cut AI costs by 50x and benefited everyone. Chamath's version: once open source is banned, you get a monopoly in partnership with the federal government.
— David Sachs / Friedberg / ChamathAI can't even nail autocomplete yet
The core anti-Doomer argument isn't theoretical speculation but everyday experience: autocomplete still isn't good, and AI can't even handle a chore like "bring the toiletry bag upstairs." If it can't finish one mundane task, why believe it can kill all of humanity? They call this humanity's insurance against AI taking over the world — company decisions still require humans to wake up, drink coffee and sit in the same room, and there is a natural 16-hour window between decisions.
— Chamath / JasonWhat's truly damning is that retweet endorsement
From a legal standpoint, the key isn't Jacob Coxon's resignation tweet but Evan Hubinger's endorsement retweet — he is still a senior alignment executive at the company, running one of the safety teams, and his co-sign plus the "greater than 10% probability" figure is what made it go viral. If it were only Coxon, the company could call him a disgruntled employee; but an executive endorsement turns "was disclosure adequate" into a real question.
— David SachsOpenAI solved the math problem with brute force, not divine inspiration
The Navier-Stokes solution used roughly 130 billion output tokens, with 10,000 agents communicating with each other, equivalent to between 50,000 and 500,000 years of human work (estimated at 40 words per minute of typing). AI had no flash of insight humans couldn't have; it just did a huge amount of brute-force work, and every step of the message chain is recorded and human-readable. The conclusion is that AI is a leverage tool, not a math god above humans.
— ChamathDe-identification doesn't remove the method itself
De-identification can strip personal information and company-specific information, but the general method for solving a math problem is itself IP. A user iterates on that method inside the tool, and that conversation can be de-identified and used for training, making the model better at math. This is the network effect of closed models: they can see what the whole world is doing, and this observation system feeds them more data than anyone else, constituting a long-term advantage over customers.
— Sacks / ChamathIt's not that the ads got worse — the shoes fall apart in six weeks
Sacks says all his shoes used to be Nike, then he stopped buying them, and the reason wasn't advertising or the ordering experience but the product itself getting worse — shoes fall apart in about six weeks, thin and not durable. He walked into REI, tried on a pair of Brooks and switched. Chamath says he stuck with Nike a long time, buying a pair every six months, then it became every six weeks, and he finally switched to On because of knee problems, for a straightforward reason: Roger Federer represents mastery and excellence. Brooks is owned by Berkshire Hathaway, has posted nine consecutive years of double-digit growth, and has $1.6 billion in revenue; its CEO says Buffett gives just one instruction a year: make the product better than it was at the end of last year.
— David Sachs / ChamathIn their own words · checked verbatim
The people building AI earnestly believe that it could kill us all by the end of the decade.
On the one hand, you're asking public market investors to underwrite your company to a value in the trillions. On the other hand, your own safety lead, not Jacob Cox in the boss who then sort of co-signed his tweet, is saying that your core product is unsolved and potentially civilization ending.
David Sachs10:18
this is a doomer psyop and the ultimate target is open source
David Sachs24:49
Autocomplete doesn't work yet, guys.
Chamath39:17
we are nowhere near removing the human from the loop of AI development
Friedberg40:20
AI is a tool of leverage for humans. AI is not some magical genius super god that sits above us, that understands things we don't understand.
Chamath1:00:12
There's a concept inside of these models called zero data retention, ZDR. It's a best efforts basis. It's a commercially best efforts basis at that. They can't guarantee it.
Chamath1:04:21
as long as we have a closed model duopoly of Frontier AI, and they're reserving the right to get into every vertical application there is, they're declaring in advance they're going to compete with their customers. So how can we trust them with our data?
Figures
| Coxon's tenure at Anthropic | six weeks | 0:00 |
| Extinction probability within a decade, per Hubinger | above 10% | 1:04 |
| Nuclear cost per gigawatt, US vs France vs China | $15 billion / $4 billion / $1 billion | 16:35 |
| Output tokens OpenAI used to solve Navier-Stokes | roughly 130 billion | 1:00:12 |
| Nike's peak market cap | $264 billion (2021) | 1:20:50 |
| Nike's peak revenue | $51 billion (2024) | 1:20:50 |
| Decline in Nike's China sales | 30%, eight consecutive quarters of decline | 1:20:50 |
| Brooks revenue | $1.6 billion, nine consecutive years of double-digit growth | 1:27:53 |
Glossary
- Doomer
- The camp that believes AI will most likely cause human extinction, often overlapping with effective altruism (EA) funding networks.
- RSI / recursive self-improvement
- The loop in which AI helps improve the next generation of AI; a mild version is already happening, while the extreme version means AI fully autonomously designs and launches the next training run.
- ZDR / zero data retention
- A clause in which a model provider promises not to retain user data, but it is only a commercial best effort and cannot be guaranteed.
- S-1
- The prospectus a US company files with the SEC before an IPO, which must accurately present both opportunity and risk.
- DTC / direct-to-consumer
- A strategy of selling straight to end users while bypassing retailers; Nike pushed it aggressively and alienated retail partners.
How to listen
Founders and investors watching AI regulation, the open-source ecosystem and IPO disclosure risk; also brand and consumer-sector people who want to dissect how Nike's brand fell apart.
The nuclear cost and Three Mile Island historical analogy (roughly 15:34–19:38) can be fast-forwarded; it is loosely tied to the main thread.