Many people have read the Anthropic prospectus in vain.
Over the past two days, everyone has been reading Anthropic's prospectus, and tens of thousands of interpretive articles have been published, with almost identical views. It is estimated that 80% of these articles are generated by AI that reads the document on behalf of humans.
After reading this prospectus word for word, we found many interesting facts that have been overlooked.
Some facts have been noticed by a few people, but they either did not interpret them, or their interpretations went in completely wrong directions.
The money spent on security for Dario is four times his annual salary.
In 2025, Dario Amodei and his sister Daniela Amodei each received an annual salary of $654,000. However, the security expenses Anthropic paid for the two were $2.746 million for Dario and $1.294 million for Daniela.
Dario's personal security expenses disclosed in the 2025 prospectus | Source: Anthropic Prospectus (Form S-1)
The total security expenses for the brother and sister, plus Dario's private jet costs, add up to $4.14 million, which is about 30 times the median security expenditure for CEOs of S&P 500 companies.
Of course, this amount is not much compared to Mark Zuckerberg, who is one of the most hated public figures, with annual security expenses exceeding $20 million. But it is very high compared to others: Dario's personal security expenses alone are equivalent to that of three Tim Cooks or eighteen Satya Nadellas.
The high expenditure is because there are indeed people who want to harm the founders of these AI companies.
In April this year, Sam Altman, CEO of OpenAI, had his home in San Francisco attacked by Molotov cocktails and shot at successively;
Surveillance footage outside Altman's San Francisco residence, the circled figure is the suspect who threw the Molotov cocktail | Source: ABC7
In the same month, someone broke into the lobby of Anthropic and threatened to kill people. In August, another person mentioned in a chat with Claude that he would use an AR-15 rifle to deal with Dario.
The model of the gun was stated so precisely that it looked extremely like a real threat.
Being a CEO of an AI company is a high-risk occupation. Many of OpenAI's security personnel have backgrounds such as FBI agents. The San Francisco Standard once reported that plainclothes guards wearing black baseball caps, sunglasses and transparent earphones are everywhere downstairs of OpenAI.
Guards on duty outside the OpenAI office, June 2024 | Source: Garrett Leahy / The San Francisco Standard
If Donald Trump had such security conditions when he campaigned in Pennsylvania in 2024, the gunman would not have even been able to climb onto that roof, and Trump would probably not have become president.
On the eve of submitting the prospectus, the brother and sister added a "dismissal insurance" for themselves.
In 2025, Dario and his sister had no severance compensation at all, which is clearly stated in the prospectus.
In October 2026, on the eve of submitting the prospectus, the company signed an agreement with them: as long as they are fired without cause, or resign voluntarily for "good reason", each of them can get 24 months of basic salary, plus 12 months of medical insurance subsidies, and their equity will vest 12 months in advance.
Prospectus: If they leave the job under agreed conditions, the brother and sister can get compensation including 24 months of basic salary | Source: Anthropic Prospectus (Form S-1)
There is no statutory termination compensation such as "N+1" in US labor law.
Most states implement the "at-will employment" rule: companies can fire employees without paying any compensation legally. Severance compensation depends entirely on the contract negotiation and the company's conscience.
Converted according to Chinese labor law, the two of them have set a compensation standard equivalent to 4N for themselves.
Why did they suddenly add such a clause? Because the company is going to be listed, Dario and his sister designed a governance structure themselves, which creates the possibility that they could be dismissed, which we will explain later.
The world's most advanced AI company has a personnel structure similar to that of private enterprises in Zhejiang, China
Anthropic is a huge Amodei universe.
The relationship between Anthropic's founding team and Dario's relatives and friends, the avatars are illustrative illustrations | Source: Arranged by Poker Table based on the prospectus, company announcements and public reports
Many people know that the older brother Dario serves as CEO and the younger sister Daniela serves as President and Chairman. The two siblings seem to have known since birth that they will work together to change the future of humanity, and they have discussed philosophical issues at home since childhood.
At the same time, the spouses of the two siblings also work in this company, so the brother, his wife, the sister and her husband all work together in the same office.
Dario's wife Cami Clark has no official position, but according to The Wall Street Journal, she is his important advisor, who accompanies him to events and meets investors. Without an employee ID, it does not prevent her from participating in major matters.
In February 2026, Dario and his wife Cami held hands and attended the New Delhi Summit | Source: Ludovic Marin / AFP / Getty Images, published by WSJ
Therefore, in Anthropic's prospectus, there is only one officially disclosed working relative: Holden Karnofsky, the husband of Chairman Daniela, that is, Dario's brother-in-law.
He has been working at the company since January 2025, and the prospectus specifically notes that his compensation is "consistent with employees with comparable experience and positions".
But Karnofsky is not just the brother-in-law, he is also Dario's former roommate, and a close comrade who promoted the effective altruism movement together.
He is the co-founder of Open Philanthropy, the flagship institution of the effective altruism movement. At that time, Dario, who was still a researcher at OpenAI, served as the technical advisor of this institution.
In Silicon Valley, networking is extremely important. Anthropic first got investment from Eric Schmidt, former CEO of Google, which was brought in by Dario's wife (his girlfriend at that time); even Dario himself got into OpenAI thanks to his sister's recommendation.
When Daniela worked at Stripe, her supervisor was Greg Brockman, who later co-founded OpenAI. When Brockman was preparing to establish OpenAI, she did not avoid involving her own relatives: "The smartest person I know is my brother."
In addition to Dario's whole family working together, there are many old acquaintances in the company.
Jared Kaplan, Chief Scientist of Anthropic, is a friend Dario met during his graduate school years, and the two once shared an apartment in San Francisco.
When Kaplan explained why he switched from physics to AI, one of the reasons was: "I want to spend more time with my friends."
We make an irresponsible guess that this friend is Dario.
The donation is very likely to fail to materialize
Dario wrote in his long article *Adolescence of Technology*: "People at the forefront of the AI wave should be willing to give up their wealth and power. All seven co-founders of Anthropic have promised to donate 80% of their personal wealth."
Seven founders of Anthropic sitting and talking around | Source: Anthropic / Forbes
The seven people almost equally share the original shares of Anthropic. According to the price of the latest round of financing this year, each person's wealth is between 14 billion and 15 billion US dollars.
If all seven people donate 80% of their wealth, the total amount will be about 84 billion US dollars, which is more than Myanmar's annual GDP (about 82.4 billion US dollars), equivalent to one fifth of the total annual personal charitable donations in the United States (394.2 billion US dollars).
However, the prospectus emphasizes that: This promise is not legally binding.
This commitment is "voluntary and revocable", and "does not constitute a legally binding obligation".
Original text of the prospectus: The donation commitment is voluntary and revocable, and does not constitute a legal obligation | Source: Anthropic Prospectus (Form S-1)
When to donate, how much to donate, and who to donate to are all uncertain, depending on personal circumstances, tax and inheritance planning, and market conditions.
Does donating 80% sound very attractive?
Actually not. At the end of 2015, after Zuckerberg's daughter was born, he announced:
During his lifetime, he will donate 99% of the shares he holds.
In 2015, Zuckerberg and his wife with their newborn daughter Max | Source: Facebook / France 24
Moreover, Zuckerberg's idea is exactly the same as Dario's: donating money cannot dilute power. Facebook proposed to issue a new class of shares with no voting rights, so that he can sell shares to donate money while retaining control.
As a result, shareholders sued the company to court, and this plan was abandoned in 2017, and the matter was left unfinished.
Zuckerberg still donates money every year through his own charity foundation, but as Meta's market value continues to rise, the amount corresponding to this 99% commitment is getting larger and larger, and the fixed annual donation is getting further and further away from the original promise.
Let's hope Dario's 80% donation promise will not end up unfinished.
Media noticed that one co-founder was isolated, but in fact this guy never got a seat at the table
Some sharp-eyed media noticed: On the day the prospectus was submitted, *The San Francisco Standard* published a report saying that one co-founder was not included in the limited liability company (Founder LLC, hereinafter referred to as "Founder LLC") jointly established by the seven founders.
In various public occasions, Anthropic calls Benjamin Mann a co-founder, listed alongside Dario and the other seven people. He is also the co-first author of the GPT-3 paper.
On October 14, 2026, Ben Mann spoke at Anthropic's event in Singapore | Source: Suhaimi Abdullah / Bloomberg / Getty Images, published by SF Standard
However, the prospectus specially wrote a footnote for him, which in plain language means: He is a founder, who left OpenAI to start a business with us back then, but the holding entity of our seven founders does not include him.
There are several possibilities for excluding Benjamin Mann:
1. Only 7 people are left in the founder holding entity, and an odd number is convenient for voting;
2. Benjamin Mann has personal stains;
3. He does not agree with Dario's ideas;
We can rule out these reasons one by one:
1. The number of voters does not matter for this Founder LLC
Odd-number voting is indeed