Remember when four of the most powerful AI companies on Earth pooled their money to make AI safe? It was July 2023. Anthropic, Google, Microsoft, and OpenAI formed the Frontier Model Forum, described as the industry's primary self-governance initiative for frontier AI safety. Three months later, they pledged over $10 million into a new AI Safety Fund. The goal? Tools for testing and evaluating the most capable AI models, plus standardized third-party evaluations. Sounds great, right?
But here's the twist: the evidence base for this story is dominated by 2023 news. Nine of twelve search hits describe the Forum's founding and the October 2023 fund announcement. That's a three-year-old story being surfaced as if it were current. I found no verifiable 2025 or 2026 development about this fund—no grant announcements, no funding updates, no governance changes. And that silence is deafening.
Why should you care? Because while these companies were writing modest checks, the rest of the AI world was racing ahead. Alibaba, ByteDance, and DeepSeek are reportedly chasing models with up to 10 trillion parameters. That's up from 671 billion just a year ago. Think about that: safety research funded at $10 million while model sizes explode past 10 trillion parameters. That's like installing a home security system while your house is being rebuilt into a skyscraper.
The critics got it right in 2023. Yahoo Finance called it throwing a 'tiny bone' to safety researchers. The sum 'might seem like a drop in the ocean compared to the tech giants' overall R&D budgets.' Andrew Maynard asked whether the projects would be radical enough. Three years later, we still don't know.
Meanwhile, the Forum's own language was telling: AI safety research is 'required.' Not urgent. Not underfunded. Required. That's the corporate equivalent of a shrug.
So here's my take: this fund was never meant to solve AI safety. It was meant to solve a PR problem. And it worked—until we forgot about it.
What happens next? Trends suggest the real safety work is happening outside these four companies, in independent labs and academic groups that don't answer to shareholders. If the Frontier Model Forum wants to stay relevant, it could either scale up dramatically or admit it was always a side project. Watch for grant announcements—or the lack thereof. The silence might tell us more than any press release ever could.