OpenAI is calling for mandatory national AI safety requirements, arguing that increasingly capable systems now justify a federal framework tied to what models can actually do.
In a policy statement published September 9, OpenAI said it wants to work with Congress on capability-based national regulation. At the same time, the company announced support for four California bills covering independent safety assessments, auditor standards, protections for young people and safeguards against AI-enabled biological threats.
What capability-based regulation means
The basic idea is to regulate systems according to measured capabilities or risk thresholds rather than applying identical rules to every AI product. A small recommendation model and a frontier system capable of advanced cyber work would not automatically face the same obligations.
That sounds simple, but implementation is difficult. Regulators still have to decide which evaluations count, who runs them, how thresholds change as models improve and what obligations are triggered when a system crosses a line.
Why the timing matters
The policy debate is moving quickly because model capabilities are moving quickly. Companies also have incentives to prefer one predictable national framework over a patchwork of state rules. States, meanwhile, have often moved first when Congress has not acted.
OpenAI's position therefore reflects both a safety argument and a regulatory-structure argument. Supporting federal rules can mean stronger national standards, but it can also shape which level of government has the most authority.
What to watch
The details matter more than the slogan. Watch how any proposed national standard defines covered models, whether independent testing is required, what disclosure obligations apply, and whether federal law would preempt stronger state rules.
Horn Updates reports the company's stated position and provides independent context. See our editorial policy.