OpenAI put out a new document this month. They're calling it the Frontier Governance Framework. It's their pitch for how they'll evaluate and constrain their most powerful models before those models get out into the world. It dropped within days of their 2026 election safety plan.
And it only binds OpenAI. Not Anthropic, not Google, nobody. None of them have to line up, and none of them are required by law to write any of this in the first place. Worse, a company policy lasts exactly as long as the people running the company want it to. New board, new chair, new priorities, and the same document quietly becomes a memo nobody opens. That's not a knock on the current team. It's just what a voluntary commitment is.
