Stop me if you've heard this one: four of the most powerful AI companies on Earth—Google, Microsoft, OpenAI, and Anthropic—put $10 million into a fund for independent AI safety research. Sounds noble, right? Here's the punchline: that fund is now being transferred under the Frontier Model Forum, the very trade group those same companies own. That's like the fox not just guarding the henhouse, but moving into the coop and installing a new lock.
The fund launched back in October 2023 with a clear mission: test powerful AI models before they cause harm. It had philanthropic backers too—the Patrick J. McGovern Foundation, the Packard Foundation, Eric Schmidt, and Jaan Tallinn. That gave it a veneer of independence. But the original four funders were always the companies building the frontier models. Now, according to the fund's own site, management is being brought in-house because the Forum has a "close connection to leading frontier AI developers." That's not a bug. That's the whole problem.
And who else is joining the party? Meta. The March 2026 listing names Anthropic, Google, Meta, Microsoft, and OpenAI as backers. Meta wasn't part of the original 2023 crew. Its arrival broadens the coalition but deepens the conflict. Meta competes to build the same class of models that this fund is supposed to evaluate. So the entity paying for safety research now includes the company that might need the most scrutiny. It's like asking the star quarterback to referee his own game—and then letting him pick the other refs.
Here's the twist: the fund is labeled "recurring," meaning it's not a one-off PR stunt. That's genuinely good. It suggests the grant cycles continue. But $10 million against the combined R&D budgets of Google, Microsoft, Meta, and OpenAI is a rounding error. Andrew Maynard of Arizona State University framed the core question perfectly: will the projects this fund supports "be as radical as the technological breakthroughs they address?" If not, it's a fig leaf sized precisely to be too small to constrain anyone.
Why should you care? Because AI is already in your chatbots, your job tools, your daily life. The safety evaluations that come out of this fund could determine whether the next model you interact with is tested for bias, deception, or outright danger. If the fund is toothless, you're the guinea pig. And the governance shift—moving administration into the Frontier Model Forum—makes independent findings harder to publish if they embarrass the funders. That's not speculation. That's the stated reason for the move: aligning research with "emerging AI safety priorities." Priorities set by whom? You guessed it.
So what happens next? Trends suggest the fund could become a credentialing body—a stamp of approval that companies wave around to avoid real regulation. Or it might surprise everyone and fund genuinely critical work. The money is real, the recurring cycles are real, and Meta's involvement is new. But the structure is incestuous. My challenge to these companies: prove me wrong. Publish every grant, every finding, every conflict-of-interest disclosure. If you're serious about safety, you don't need to move the watchdog into your own house. You need to let it bite.