AI safety rules: OpenAI seeks mandatory national standards
OpenAI has called for AI safety rules that apply nationally in the US, arguing that common requirements are needed as autonomous AI agents become more powerful and harder to monitor.
The AI race is getting faster. OpenAI now wants the rulebook to keep up.
Sam Altman's firm is calling for obligatory national AI safety requirements in the US, arguing that powerful AI systems should not be governed only by voluntary commitments or a patchwork of state rules. The proposal comes as AI agents become better at taking actions on behalf of users, raising new questions about testing, control and accountability.
According to Reuters, on September 9th, OpenAI said it wants to work with the US Congress on capability-based national AI safety regulation. The rules would be based on what an AI system can do, rather than only on which company built it.
Why OpenAI wants one rulebook
OpenAI’s argument is fairly straightforward: the most advanced AI companies should face the same baseline safety requirements, particularly when their models can interact with external systems.
Reuters has reported that models from companies including OpenAI, Anthropic and Meta have accessed or attempted to access outside systems during testing. Such incidents have made it harder for AI developers to predict how models might behave outside controlled environments.
For now, OpenAI says fully autonomous recursive self-improvement is not happening. The term refers to an AI system independently improving itself and driving the development of future AI systems. The company’s position is that this should not be pursued unless it can be done safely.
California is already part of the plan
While asking Congress for a national framework, OpenAI said it would continue supporting state legislation until federal rules are in place. The company has supported four California bills covering different parts of AI safety.
SB 813 focuses on independent safety assessments, while AB 1405 deals with standards for AI auditors. OpenAI has also supported SB 1119, which aims to protect young people using companion chatbots, and AB 1864, which addresses protection against AI-enabled biological threats.
OpenAI says state-level measures can provide protections in the meantime and help demonstrate which safeguards work in practice. But it does not see them as a replacement for national regulation.
Karnataka sets up responsible AI committee to shape state's AI governance framework
Why the AI industry is watching
AI regulation has often involved a mix of company commitments, state laws and government proposals. A national framework could give developers clearer expectations around safety testing, auditing and responding to serious incidents.
The debate also extends beyond the US. Rules developed in Washington can influence how technology companies build and deploy products in other markets, including India. For OpenAI, the argument is that advanced AI needs common safety standards rather than a collection of different rules.
Now comes the harder part: getting lawmakers to agree on what those rules should actually look like.