Ethereum co-founder Vitalik Buterin recently disclosed in detail for the first time in a podcast interview that the Shiba Inu (SHIB) tokens he donated to the Future of Life Institute (FLI) in 2021 ultimately cashed out for roughly $400 million to $600 million, making it the largest single donation the organization has received to date. But the two sides have developed a fundamental disagreement over how the funds should be used: FLI has shifted from focusing on all existential risks facing humanity to betting entirely on AI safety and deeply engaging in policy advocacy, and Buterin's repeated requests to broaden its focus went unheeded.
The origin of this donation was rather dramatic. In 2020, the Shiba Inu project team sent half of the token's total supply directly to Buterin's wallet. He initially sold off a portion quickly for Ethereum, cashing out about $70 million, which he mainly donated to the Against Malaria Foundation to buy mosquito nets. Because of insufficient market depth, the remaining large amount of Shiba Inu tokens could not be easily liquidated, so he simply donated half of them to India's pandemic relief organization CryptoCOVID Relief and gave the other half to FLI. By his estimate at the time, the latter would receive at most a little over $20 million. Three months later, the token price did not collapse as expected: CryptoCOVID Relief cashed out $470 million, while FLI's cash-out scale landed in the $400 million to $600 million range.
Buterin admitted that this outcome was completely beyond imagination, and that had he known, his donation arrangements would have been entirely different. He later persuaded CryptoCOVID Relief to return half of the funds, which he used to independently advance pandemic relief projects, including air filters, ultraviolet light, masks, long COVID research, and funding for the open-source nasal spray vaccine project RADVAC and the Indian biotechnology company PopVax. According to him, about $200 million has already been deployed in substantial work across these areas.
The turning point in the disagreement came after the release of GPT-3. Buterin said FLI's focus narrowed from biological, AI, nuclear war, and other existential risks to the single issue of AI, and its approach shifted from research to political action, beginning to judge which bills and politicians to support. He communicated two or three times requesting an expansion of its scope, but was refused each time. This pivot also triggered backlash in Silicon Valley investment circles, with some investors viewing the AI safety agenda as an opposing force.
Buterin used a "human vs. nature" and "human vs. human" framework to explain his reflections. He believes Effective Altruism (EA) is extremely effective in "human vs. nature" domains such as fighting disease and improving public health, because donors only need to compare how many lives different organizations can save. But once it enters "human vs. human" domains such as policy contests, judgment becomes far more difficult—the effects of policies are hard to predict, and large sums of money openly intervening in political issues may provoke public resentment, thereby weakening the very support base for the AI safety agenda itself.
His subjective probability estimate of the risk that AI destroys humanity is 0.1, but he stressed that this number overlays three layers of uncertainty: technological pathways, the alignment problem, and geopolitics. AI development could be as fast as ten years or as slow as three hundred years; whether the alignment problem is truly as severe as rationalist Eliezer Yudkowsky claims remains undetermined; and in a landscape where major powers and multiple companies are racing toward AI, once external conflicts escalate, the last thing any party will prioritize is safety.
Regarding the debate over Anthropic, Buterin summarized it as two camps within EA. Supporters argue that since the AI race cannot be stopped, the company upholding safety principles should win, and that Anthropic's investment in safety research is superior to OpenAI's. Opponents criticize Anthropic's participation as itself accelerating the race, and argue that its safety philosophy implicitly carries the logic of "we hold power because we are the good guys," with its connection to the company's own interests possibly exceeding 90%. He noted that Anthropic's opposition to open source earlier this year further intensified this debate.
As an alternative path, Buterin proposed d/acc (defensive accelerationism), which rests on three pillars: acknowledging the importance of technological development; prioritizing resources toward protective technologies such as cybersecurity, vaccines, and air filtration; and adhering to decentralization and open source, empowering ordinary people rather than concentrating power. Using pandemic prevention as an example, he pointed out that some EA organizations bet 80% of their resources on restricting AI models' ability to generate dangerous viruses—a centralized defense that relies on the assumption of "whoever holds the most powerful model." The decentralized path, by contrast, focuses on faster detection, better filtration, and faster vaccines, which can work even without controlling all AI.
The interview also touched on the overall situation of the EA movement. According to tech journalist Afra, EA has gone through four stages: the "mosquito net period," the longtermism period, the reputation-collapse period, and the current AI safety period. The collapse of FTX founder Sam Bankman-Fried dealt the movement a crisis of trust, while the rise of the AI safety issue has pushed its influence to new heights. She noted that regarding the funding sources of METR, an independent safety audit organization tied to Anthropic, outsiders have questioned whether its funders overlap with Anthropic's early investors, but she views this as more of a conspiracy theory, since EA's agenda is not hidden.
Worth noting is that EA's deep entanglement with policy circles has triggered political backlash. The U.S. Department of Defense once publicly stated on social media that it was distancing itself from EA, and the Trump administration has also been drawn into related debates. Afra pointed out that traditional Washington think tanks now struggle to recruit qualified AI policy analysts, with young talent flowing in large numbers to EA-funded institutions. This dual concentration of talent and funding is reshaping the agenda-setting power over AI governance issues.
Buterin's ultimate judgment on AI safety carries a degree of openness. He believes AI safety is less an engineering problem than a political design problem—drawing on institutional architectures that combine checks and balances with communication and isolation, using multiple imperfect but mutually constraining mechanisms to reduce systemic risk. But he also stressed that the speed and direction of AI development over the past decade have repeatedly defied expectations, and that excessive certainty about any path at this moment is inadvisable; maintaining an open mind is key.
For the crypto market, the subsequent handling of this donation also offers an observation sample: the actual cash-out scale of large token donations depends heavily on market liquidity, donors find it difficult to plan in advance, and the final destination of charitable funds may deviate significantly from original intentions. For the field of AI governance, when hundreds of millions of dollars in private funds simultaneously act on technology research and policy lobbying, funding sources, institutional independence, and agenda transparency will become focal points of sustained scrutiny.
