In 2004, Oxford University received the single largest donation in the university’s history, a sum of £60m from information technology engineer James Martin. Martin developed an approach to software development called “rapid application development”, which involves truncating the planning phase of app development and instead using iterative development of software prototypes that you can break and play with to build the best tools. His approach is now dominant in the development of user interfaces, and it earned him recognition as the #4 most influential person on computer science in Computerworld’s 1992 25th anniversary issue. 

But Martin’s biggest influence on computer science wasn’t his lifetime of work improving software engineering methodology. It was the 2005 founding of The James Martin 21st Century School, better known today as the Oxford Martin School. The school aimed to “formulate new concepts, policies, and technologies that will make the future a better place to be”, and its first major project was to establish and fund the Future of Humanity Institute, a multidisciplinary research group founded by Professor Nick Bostrom. FHI’s early staff included characters like Swedish futurist Anders Sandberg — a man who manages to simultaneously be at the center of transhumanism, cryonics, human enhancement, and the search for extraterrestrial intelligence — and Eric Drexler, the earliest pioneer of nanotechnology. Together they conspired to build a better future for humanity, assessing the social and technological factors that could positively and negatively shape our future. 

2014 saw the publication of Bostrom’s seminal work, Superintelligence. The book was the first widely read strategic analysis of superintelligence risks and the AI alignment problem. It was recommended by Bill Gates and by subsequent OpenAI co-founders Elon Musk and Sam Altman. 

Bostrom’s Superintelligence and his subsequent work advising think tanks, AI labs, and governments can very credibly be said to have founded the fields of AI alignment and AI safety, fields that would later become a core part of the strategies of frontier companies OpenAI, Anthropic, and DeepMind.18 But this was only one of FHI’s numerous contributions to safer AI. 

Throughout its 19-year tenure, FHI developed the idea of AI governance and came up with the foundational ideas that made the field possible, and counted among its alumni two of OpenAI’s governance leads, Google DeepMind’s governance and safety lead and its biosecurity lead, Anthropic’s policy lead, senior leadership at RAND Corporation, C-suite leadership at the AI Security Institute (the world’s most seminal government institution for secure AI), and grantmakers for major foundations that would allocate hundreds of millions of dollars to AI safety, defensive AI, and worst-case biosecurity. It developed the idea of a “global catastrophic risk” and led to the publication of Toby Ord’s best-selling book The Precipice, which would be read by some of the world’s highest government offices. 

By being far ahead of the curve on artificial intelligence, the Oxford Martin School’s Future of Humanity Institute would not so much shape the fields of AI safety, security, and governance, but bring them into existence. There is a credible case that the relatively good safety culture of AI labs — staffed with safety, alignment, and preparedness teams and with C-suite leadership who talk openly about these problems — and the relative strategic clarity that we have about AI safety (e.g. see Section IV) are a direct result of the Oxford Martin School. 

Bostrom and FHI have had their detractors, and some of the founder’s behavior would unfortunately come to mirror the behavior of other early adopters and mad scientists, who manage to be tremendously seminal through their idiosyncratic, and at times reckless, behavior. But whatever its flaws, there can be zero doubt that through Bostrom and FHI, legendary computer software engineer James Martin had his last and greatest impact on the future of technology. 

From 2014 through 2024 about a half a billion dollars went into AI safety philanthropy. Funders of this work have included James Martin, Dustin Moskovitz, Eric Schmidt, Vitalik Buterin, Elon Musk, Jaan Tallinn, Bill Gates, the David and Lucile Packard Foundation, the Omidyar Network, TED Audacious, Amlin, the European Research Council, the National Science Foundation, and the Leverhulme Trust. 

Half a billion dollars, however, is not a lot of money. For comparison, it is 1/1,000th of Project Stargate, the new computing infrastructure project announced by OpenAI, Softbank, and Donald Trump in early 2025. Microsoft invests more than a hundred times this much money into OpenAI every year for its data centers. Seven Anthropic co-founders have individual net worths that are seven multiples of this figure. 

Yet when compared next to the contributions of the top eight AI companies, which, as of one year ago, collectively controlled $120 billion,19 and compared to the contributions of governments, with budgets in the trillions, the impacts of philanthropy’s shoe-string budget look astounding. This is in some part because philanthropists have been so far ahead of the curve. And today we are still ahead of the curve on the issue of AGI, with many public commentators denying it entirely, and markets failing to anticipate AI’s transformative impacts. Further, today there is vastly more strategic clarity about AI safety, security, and preparedness than at any other moment in history. We can have more impact now than the early philanthropists who defined the field. 

To date, philanthropically-funded work has (1) started the field of AI safety and several subfields, including governance, control, corrigibility, interpretability, robustness, and evaluations; (2) substantively shaped the most important legislation on AI; (3) developed the most central AI development and deployment policies of AI companies; (4) trained many of the most central figures in the field and hundreds of researchers; (5) started the only dialogues between the U.S. and China on AI safety; (6) cultivated the top talent in AI security in Washington; and (7) driven the most widely-read communications campaigns in AI safety. 

Below I detail some of the biggest wins from nonprofit organizations (FHI excluded) for safety and security over the past decade: