Anthropic IPO insanity: the ex-housemates who shaped AI's safety checks

By Business & Technology Global · Thu Nov 12 2026

Next week Anthropic plans to sell 825 million shares at $70 to $78 each, a potentially historic Nasdaq debut that could value the Claude maker at up to $1.405 trillion. In the prospectus, its co-founders say their aim is to "help humankind chart a safe course" through AI. Two of the men who helped build AI's safety checks, Holden Karnofsky and Paul Christiano, once lived under the same roof as Anthropic's chief executive.

The 2017 footnote

In March 2017, Open Philanthropy published its reasoning for a recommended $30 million grant to OpenAI, with a short section headed "relationship disclosures". Two OpenAI researchers, Dario Amodei and Paul Christiano, lived in the same house as Holden Karnofsky, the foundation's executive director. Karnofsky was also engaged to Amodei's sister, Daniela.

Nine years later, all three hold some of the most consequential jobs in AI. Amodei runs Anthropic. Karnofsky, now his brother-in-law, led the latest rewrite of its safety rulebook. And Christiano, who left Anthropic's governing trust in 2024, sits on the board of the foundation that controls OpenAI.

The 2017 grant page also described the partnership that came with the grant. Karnofsky would join OpenAI's board and, with one other director, oversee its safety and governance work.

That small circle keeps turning up in the paperwork. It is the hook of a viral video posted on X on Thursday by app developer Mo (@atmoio), whose post passed 50,000 impressions within hours. His question for would-be Anthropic shareholders: who actually built the "independent" checks on the company?

Holden Karnofsky: the Bridgewater alumnus who set out to grade charities

Karnofsky, born in 1980 or 1981, was a National Merit Scholar at Stevenson High School in Lincolnshire, Illinois. He studied social studies at Harvard, where he wrote for the Harvard Lampoon, and graduated in 2003. He then joined Bridgewater Associates, the giant hedge fund.

At Bridgewater he and colleague Elie Hassenfeld started a club to research which charities actually worked. In 2007 they quit to run it full time as GiveWell.

That December, Karnofsky posted a question about charities on the forum MetaFilter under another person's name, then answered it as himself, recommending GiveWell without disclosing that he ran it. He stepped down as executive director and was later reinstated; the board cut $5,000 from his salary to pay for a professional development course. The New York Times covered the episode at the time.

GiveWell survived and found a patron in Good Ventures, the foundation of Facebook co-founder Dustin Moskovitz and his wife Cari Tuna. Their joint "GiveWell Labs" grew into Open Philanthropy in 2014, with Karnofsky in charge. By its own count, the group has directed more than $4 billion since 2014.

In 2021 he published "The Most Important Century" on his blog, Cold Takes. The series argues that advanced AI could make this century the most consequential in human history. He stepped back from running Open Philanthropy in 2023, spent 2024 as a visiting scholar at the Carnegie Endowment for International Peace and joined Anthropic in January 2025 as a "member of technical staff".

His primary focus there, he has written, was the Responsible Scaling Policy, Anthropic's voluntary framework for mitigating catastrophic risks from AI. Version 3.0, released in February 2026, dropped the implied promise that Anthropic would pause on its own and recast it as recommendations for the whole industry. "I take significant responsibility for this change," Karnofsky wrote at the time.

Paul Christiano: the math olympian who helped teach AI to take feedback

Christiano went to the Harker School in San Jose, California, and won a silver medal for the US at the 2008 International Mathematical Olympiad. He earned a mathematics degree from MIT in 2012 and a PhD from UC Berkeley in 2017, under computer scientist Umesh Vazirani.

At OpenAI from 2017 to 2021 he led the language model alignment team. He was the first-listed author of the 2017 paper "Deep Reinforcement Learning from Human Preferences", co-written with Dario Amodei and others. He is considered one of the principal architects of RLHF, in which human judgments of a model's answers are used to train it.

He left in 2021 to found ARC, a nonprofit working on theoretical alignment, including how to get at what an AI model actually "knows". ARC's model-testing team spun out in December 2023 as METR. In April 2024 he became head of AI safety at the US AI Safety Institute, now the Center for AI Standards and Innovation, after VentureBeat reported that some staff objected to his effective altruism ties. In 2023 he told the Bankless podcast there was a "10–20% chance of AI takeover" with many or most humans dead.

In August 2026 he returned to run ARC as executive director. He still advises the government part-time. In September he joined the OpenAI Foundation board and its Safety and Security Committee, which TechCrunch reports has the final say on whether OpenAI releases new models. OpenAI says he recuses himself from all OpenAI matters and model evaluations in his government role.

Same names, different paperwork

Follow the documents and the same two names keep appearing.

• ARC's tax returns: Its Form 990s for 2021 and 2022 list Christiano as president and Karnofsky as secretary and treasurer. The 2023 filing says Karnofsky left ARC's board in April 2023.
• Anthropic's trust: Christiano was an initial trustee of Anthropic's Long-Term Benefit Trust, which the IPO filing says "generally has the power to elect and remove a majority" of the board. He stepped down in 2024.
• Anthropic's IPO filing: A section headed "Family Relationships" names Dario and Daniela Amodei as siblings and says "there are no other family relationships among any of our executive officers or directors." That is accurate, because Karnofsky is neither. He appears elsewhere, under "Employment Arrangements with Immediate Family Members of Our Executive Officers".

That second section says Karnofsky received non-equity pay of $430,213 in 2025 and $428,801 in 2026 through October, on top of his equity awards. The filing also says he sold Anthropic shares in an April 2026 tender offer. Anthropic says his compensation was in line with comparable employees.

Mo's video argues these "independent" checks form a closed loop. What the record shows is narrower: each tie above is documented in public filings, announcements or reporting.

Quick facts: side by side

Why it matters before the IPO

At the top of that range Anthropic could be worth up to $1.405 trillion on a fully diluted basis, CNBC reported. Buyers are being asked to back a company whose stated purpose is "the responsible development and maintenance of advanced AI for the long-term benefit of humanity". Its latest safety rulebook rewrite was led by the president's husband. And its main rival's safety committee now includes Christiano, who once shared a house with Karnofsky and Dario Amodei.

None of this is secret: every tie here is documented in filings, announcements or reporting. Both men have spent years arguing publicly that the world is not ready for advanced AI. But before buying into a company that says it wants to chart a safe course, it helps to know that some of the people doing the overseeing once shared a house.

Read on NewsPlatform · More stories