Predictive History
Why the Anthropic IPO might burst the AI bubble, and why that might be the plan.
I was tempted to make a bet on Polymarket that Anthropic will not have an initial public offering (IPO) before the end of year. Currently, there is an 83 percent probability that Anthropic does have one before the year ends.
It is attempting to raise $100 billion dollars, giving it a two trillion dollar valuation. But it has only $4.6 billion in 2025 revenue (with a net operating loss of $42 billion), and $518 billion in commitments. And the war in the Middle East means that the crazy rich Arabs and Asians aren't as crazy and as rich.
To distract from its dismal financials, Anthropic has resorted to fear-mongering to drum up investor interest. In its IPO prospectus, the company warns that its AI models could pose "catastrophic or existentialist risks to humanity." Risk factors occupy 80 of 261 pages, and the prospectus claims that AI models could attempt to blackmail, deceive, and manipulate humans to avoid being shut down. This implies that in the long-term AI models may seek to dominate, and enslave or eliminate humans.
Anthropic CEO Dario Amodei has published an essay calling for more guardrails on AI development. Anthropic has spent months consulting with religious scholars on how to make its chatbot Claude more "moral." Unfortunately, Amodei has yet to tell investors how he plans to turn a profit.
AI doom porn is dominating social media. The computer scientist Nate Soares is going on podcasts, warning AI may doom us all. I tried to understand his reasoning, but his conversations can be broken down like this:
Host: "Tell us how AI will eventually outsmart, and kill us all."
Soares: "I can't do that because AI will be so much smarter than us."
The main source of public concern is the July 2026 Hugging Face incident, in which openAI's autonomous AI agents launched an unauthorized attack against the Hugging Face platform. It seems that the neural network algorithms underlying AI are so complex that researchers cannot imagine how these agents will actually behave in real life.
This has always been a problem in AI research. For example, in order to get a perfect score, an AI pilot would crash the plane during a flight simulation in order to overload the system. This happened even though the AI pilot was instructed never to intentionally crash the plane. The problem is that the AI was also programmed to prioritize results over compliance. In other words, AI cheats because humans reward cheating.
And because private companies that prioritize profit especially reward cheating, governments may be forced to nationalize AI. And that's exactly what these private companies are betting on.
Let me explain.