Frontier Model Forum Taps Brookings Vet, Launches $10M Safety Fund

The Frontier Model Forum—the industry body founded by OpenAI, Anthropic, Google, and Microsoft—just got its first executive director and a fresh $10 million to throw at independent AI safety research. Chris Meserole, formerly the director of the Artificial Intelligence and Emerging Technology Initiative at the Brookings Institution, steps into the top job. His mandate is straightforward but sticky: figure out how to safely develop and evaluate the most powerful AI models on the planet before their potential risks outpace our ability to understand them.

The appointment comes with a financial shot in the arm. Forum members, alongside philanthropic heavyweights including the Patrick J. McGovern Foundation, the David and Lucile Packard Foundation, Eric Schmidt, and Jaan Tallinn, have committed over $10 million to seed a new AI Safety Fund. That cash is earmarked for independent researchers—academics, startups, research institutions—who want to build better model evaluations and red-teaming techniques. The Meridian Institute will administer the fund, with an advisory committee stacked with external experts, AI company reps, and grantmaking veterans to steer the ship.

Meserole didn’t mince words about the tension at the heart of his new gig. “The most powerful AI models hold enormous promise for society, but to realize their potential we need to better understand how to safely develop and evaluate them,” he said. The fund directly ties back to voluntary commitments the same four companies made earlier this year at the White House, specifically the pledge to open up their systems to third-party discovery of vulnerabilities. That’s a promise that sounds good on paper, but without independent cash and a structured review process, it’s just PR.

What’s genuinely interesting here isn’t just the dollar amount—$10 million is a rounding error for these firms—but the structural signal it sends. The Forum also released its first technical working group update on red teaming, a move toward normalizing shared safety vocabulary and practices across rival labs. If this actually produces standardized evaluation tools that the wider research community can use, it might chip away at the opacity that currently defines frontier model development. Or it could just be a well-funded venue for coordinated self-regulation that keeps real scrutiny at arm’s length. The call for proposals drops in the next few months, and who gets funded—and what they’re allowed to test—will tell us which way this thing is really leaning.