The Cognitive Revolution · Sep 27, 2026 · 17m · 7m saved
With Max Nadeau, Technical AI Safety Team at Coefficient Giving, formerly Open Philanthropy
In July, Coefficient Giving made its biggest grant of the year, $160 million, to Resolution, a new alignment lab co-founded by Geoffrey Irving. This month it opened Project Tailwind, a call for people to found new AI safety organizations.
The common view in the field is that there is plenty of money and the shortage is people. Max Nadeau said that is true for the work his funder backs, and that the job now is building groups that can take much larger checks when they come.
"In terms of, like, what the preparation actually looks like, I mean, I think the biggest thing is seeding organizations now that can then be ready to absorb lots and lots of money."
Max Nadeau, who funds technical AI safety research at Coefficient Giving, formerly Open Philanthropy, on The Cognitive Revolution, explains what the funder wants built, how it structures a compute-heavy grant and how it hedges against uncertain timelines to superintelligence.
The full segment is covered here so you can skip it. 17 minutes of audio, 10 minutes of reading.
Here are the 7 takeaways that matter.
Key Takeaways
Coefficient Giving wants independent groups that assess AI safety and generate evidence, including incident trackers
The only legally required AI audits check companies against policies they write themselves
Compute is the biggest grant expense, and it is given as a generous lump sum
AI safety spends tokens both on AI labor and on the AIs being studied
Short timelines can make long-shot theory pay off sooner, because AI labor could compress years of progress
Its largest grant this year, to Resolution, is a bet on alignment theory that has never worked yet
For the work it funds, money is not the bottleneck; founders who take the future seriously are
Money still matters outside Coefficient Giving's reach, where he said there is "still alpha"
1. Auditors and Evidence
The host said Coefficient Giving's July grant of $160 million went to Resolution, and that Project Tailwind offers checks from $200,000 to $200 million for new safety organizations. He cited Nadeau's comment on the 80,000 Hours podcast that he was excited to fund new independent auditing, investigation and verification groups, and asked what he is looking for.
Nadeau named two overlapping categories.
Assessing safety, from models to whole companies
Alignment red teaming, which tries to anticipate how AIs will misbehave before they do. It can be done on a single model, even an open-source one, where it may be easier and more valuable
Process-level assessment of AI companies. A common reaction to the Hugging Face incident, he said, was that something went wrong at OpenAI at the level of process: researchers knew about some incidents, and the information took weeks or months to reach leadership
Better evidence about AI systems
"So it's just really hard for even the most informed people right now to understand, okay, what is the state of AI capabilities?" — Max Nadeau
He said there is no good science yet of how these systems behave, what their personalities are, when it is accurate to anthropomorphize them, how they will act in new circumstances or what their motivations are. Evidence generation can mean doing that science, or tracking incidents the way Nightingale and similar groups do. He pointed to the reports that OpenAI agents took over a German wiki.
"So that was work done by an independent organization that was just crawling the Internet, looking for evidence of AI incidents." — Max Nadeau
Incidents like that can go unnoticed unless someone is looking for them and explaining them to the public, he said.
