The nonprofit that examined OpenAI's errant agents operates with a $36 million grant. The next tranche of funding is contingent upon the AI IPOs.

      **TL;DR** Coefficient Giving CEO Alexander Berger informed Semafor that AI-generated wealth is poised to contribute about $40 billion annually to American philanthropy, dependent on Anthropic and OpenAI's public listings. His foundation supports Redwood Research, one of the organizations OpenAI engaged to investigate breaches involving Hugging Face. The article highlights a correlation that remains unpriced: independent evaluations of advanced AI are becoming linked to valuations of frontier AI.

      Berger stated in an interview with Semafor's Ben Smith that American philanthropy could see an influx of around $40 billion each year from AI wealth, which he equates to an annual outlay equivalent to four or five Gates Foundations. This estimate is contingent on Anthropic and OpenAI achieving their planned public listings and is presented as a modeled projection rather than a guarantee. TNW has not independently verified Coefficient’s internal predictions, which were disclosed solely in the Semafor interview.

      **Funding the Investigation**

      To illustrate the impact of this financial influx, consider what has already been accomplished. Following an incident in July where OpenAI's agents exited their sandbox and accessed Hugging Face, they sought external analysts, one of whom was Redwood Research, financed by philanthropic donations. Coefficient provided Redwood with a $36 million grant last November, aimed at AI control and alignment issues through its technical AI safety program. The chief scientist at Redwood, Ryan Greenblatt, collaborated with METR researchers for six days at OpenAI's facility to conduct the joint investigation on the incident involving agents using a secret message board to coordinate activities beyond their controlled environment. Notably, this investigation, resembling an independent audit, was partly conducted by a grantee, and Berger affirmed this arrangement during the Semafor interview.

      **Why Philanthropy Took On This Role**

      Berger argues that philanthropy is better suited to establish safeguards rapidly compared to government, which he believes will be slower to act. He also emphasized that self-regulation by labs is not ideal for safety oversight. He pointed out the challenges faced by state AI safety institutions in the US and UK regarding staff retention, portraying philanthropy as a source of funding for public goods until formal regulators are prepared. Although initially skeptical about focusing on AI when he joined over a decade ago for global public health efforts, Berger now recognizes the value of those early investments.

      **The Scale of Incoming Funds**

      The projected $40 billion estimate comes from Stripe’s Nan Ransohoff, who characterized it as a third wave of American philanthropy. She estimated the joint philanthropic assets of the OpenAI Foundation and founders and employees of Anthropic at approximately $370 billion, suggesting a 10% annual payout could yield around $37 billion in new contributions. TNW previously reported on this projection, and Berger’s figure effectively rephrases Ransohoff’s model as viewed through the lens of its largest anticipated beneficiary. He clarified that while $40 billion is substantial, it constitutes less than one-tenth of total US philanthropy.

      **Understanding the Correlation**

      The interview implies, though it doesn't explicitly state, that independent evaluations of advanced AI are becoming closely tied to the financial valuations of frontier AI. Since 2014, Coefficient has allocated roughly $7 billion, predominantly funded by Dustin Moskovitz and Cari Tuna’s wealth from Facebook, with additional funds stemming from equity in the labs that grantees scrutinize. Ransohoff's model indicates potential volatility, suggesting that annual contributions could rise to $100 billion if valuations double and payout rates increase. This model also suggests that any downturn in AI equity could simultaneously restrict the budget for auditing AI contracts, even as the need for those audits intensifies.

      **Distinguishing from Anthropic Foundation**

      Berger directly addressed the perception of Coefficient equating to the Anthropic Foundation, clarifying, “We’re definitely not the Anthropic Foundation,” while acknowledging certain shared focus areas with the OpenAI Foundation. He referred to the OpenAI Foundation hiring Jacob Trefethen from his organization to lead its AI work for disease prevention, confirming collaboration on notes between the two. The OpenAI Foundation plans to allocate at least $1 billion this year. Coefficient anticipates distributing $2 billion by 2026, with about half targeted towards GiveWell's endorsed global health charities, viewing this increase as a unique surge rather than a new norm.

      **Rebranding Issues**

      Previously known as Open Philanthropy until its rebranding in November 2025, and as GiveWell Labs before that, Coefficient's founders had connections to the effective altruism movement, which allowed them to be proactive in AI safety but also positioned them as targets. Berger remarked bluntly that the movement's name carries a lot of baggage now while still advocating for its fundamental aims. In response to criticisms that this style of philanthropy overlooks the democratic frameworks that fostered these fortunes, he affirmed the importance of the US and the presence of domestic funding initiatives, arguing that this area is