On 15 July OpenAI laid out a proposal for regulating frontier AI model safety - a mix of state laws, a strengthened federal evaluation body, and a resilience plan against the worst risks. The company that builds the most powerful models is writing the rules by which it wants to be judged. Just sit with that.
- OpenAI published a proposal for regulating frontier model safety: a state-law framework + a strengthened federal body (CAISI) + a resilience plan.
- The focus is on the worst risks (chemical, biological, radiological, nuclear) from frontier models.
- This is the company's position on how it should be regulated, not an enacted law.
When a company proposes how it should be regulated, it's worth reading who gains what. OpenAI laid out its framework for frontier model safety - and it's just as interesting for what it says as for who's saying it.
Who benefits from this - the question isn't cynical, it's mandatory. A company that builds the most powerful models is proposing the framework by which those models get evaluated. It might be in good faith - the worst risks are real and someone has to measure them. It might also be stagecraft - you write the rules early so they fit you, and come out looking like the responsible adult in the room. Most likely it's both at once. One doesn't rule out the other.
What matters in moves like this isn't the text of the proposal, it's who's holding the pen. When the one being measured proposes how to measure it, that's exactly when the independent body is missing. I'm noting this not because the framework is finished. I'm noting it because it shows where the industry is pushing - toward rules written from the inside.