Frontier Model Forum

OpenAI announced this week that it's forming a new industry group aimed squarely at the riskiest edge of AI development: the so-called frontier models, the systems that push past what came before in capability and, often, in unpredictability. The stated goals sound modest on paper — safety research, best practices, information sharing with policymakers — but the subtext is louder than the text itself.

This isn't OpenAI going it alone. The effort brings together the handful of companies actually building models at this scale, the ones with enough compute and enough ambition to matter in this conversation. That's a small club, and it's notable that its members are choosing to coordinate rather than just compete. Frontier AI has been treated as a zero-sum race for two years now; a shared safety body is the industry admitting, at least on paper, that some problems don't get solved by whoever ships first.

The timing isn't accidental. Governments in Washington, Brussels, and London have spent 2023 scrambling to figure out how to regulate systems most lawmakers don't fully understand. By building a forum that can hand policymakers pre-packaged research and proposed standards, these companies get a seat at the table before rules get written without them. That's not necessarily cynical — genuine safety research benefits from pooled resources — but it's also a classic move: shape the guardrails before someone else shapes them for you.

What's missing from the announcement is anything resembling teeth. There's no mention of independent audits, no enforcement mechanism, no outside seat for academics or civil society groups who've been asking for exactly that kind of oversight. Information sharing among the labs that already talk to each other constantly isn't the same as accountability to the public that has to live with what they build.