OpenAI wants the industry to agree on how AI gets tested and judged, not just how it gets built.
OpenAI published a call for coordinated, shared global standards covering how AI systems are evaluated, how results get reported, and how the technology is governed as it grows more capable. The company frames this as necessary groundwork for what it calls the next phase of AI development. It argues that company-by-company, ad hoc safety practices will not scale as models get more powerful and more widely deployed. The piece makes the case for coordination itself rather than laying out specific metrics, thresholds, or a governing body.
Whoever writes the first draft of a testing and reporting standard tends to shape every rule that follows it. That is true of building codes and it is true of AI: if OpenAI's framework becomes the template regulators reach for, the company that builds the models also ends up defining how those models get judged.
Industry calls for coordination are easy to publish and hard to enforce - this one arrives with no metrics, no commitments, and no one on the hook if nobody signs on.