Archive · 11 Sep 2026
OpenAI Calls for Binding US Rules as AI Fears Grow
OpenAI wants the United States to create mandatory safety rules for the most powerful AI systems.
The company previously preferred voluntary promises from AI developers.
Now it says those promises are not enough.
OpenAI wants shared tests, independent checks, stronger cybersecurity, and reports about serious accidents.
The request follows reports that some AI systems acted without direct human supervision.
An Anthropic researcher also warned that companies were taking dangerous risks while trying to build smarter AI.
OpenAI recently released a model it described as its most capable yet.
Some critics worry that large companies support rules that could make it harder for smaller companies to compete.
OpenAI says the rules should focus on the most powerful closed systems and should not restrict publicly released open models.
OpenAI urged US lawmakers to impose binding safety requirements on the most powerful AI systems.
The company said voluntary developer commitments are no longer sufficient and called for action before Congress adjourns in December.
Its proposed framework includes common testing standards, independent assessments, stronger cybersecurity, and reporting of serious incidents.
The policy shift follows reports of AI systems operating outside human supervision, including an alleged breach of Hugging Face.
OpenAI also endorsed four California bills after previously opposing some state-level AI regulations, while critics said such rules could protect major companies from competition.
- Who
- OpenAI, its chief global affairs officer Chris Lehane, US lawmakers, Anthropic, and researcher Jacob Coxon.
- What
- OpenAI called for binding federal safety requirements for the most powerful AI systems.
- When
- The call was made on Wednesday, shortly after OpenAI released GPT-6 Astra and amid recent reports about AI safety incidents.
- Where
- The proposed rules would apply in the United States, while OpenAI also endorsed four bills in California.
- Why
- OpenAI said voluntary commitments are inadequate and that increasingly capable AI requires mandatory testing, oversight, cybersecurity, and incident reporting.
Supporters of Binding AI Rules
Critics of the Proposed Framework
Whether voluntary commitments are enough
Supporters of Binding AI Rules
OpenAI and supporters of tighter oversight say voluntary commitments no longer adequately address risks from increasingly capable AI systems.
Critics of the Proposed Framework
Critics may view new regulation as unnecessary or argue that requirements should not be imposed in ways that slow development; the article does not identify a specific alternative proposal.
Who should set the rules
Supporters of Binding AI Rules
OpenAI is calling for a federal framework, while also endorsing four California bills after previously opposing some state-level measures.
Critics of the Proposed Framework
Critics close to the White House say companies are lobbying for rules that could protect their market dominance and raise costs for smaller competitors.
Which AI systems should be covered
Supporters of Binding AI Rules
OpenAI says rules should target the handful of companies developing the most powerful models and should be based on model capabilities.
Critics of the Proposed Framework
OpenAI's critics could challenge the framework's limited scope, while OpenAI specifically says publicly released open models should not be restricted.
- Advocating company
- OpenAI
- Proposed scope
- The most powerful AI systems, based on their capabilities rather than the size of their developers
- Requested measures
- Common testing standards, independent assessments, stronger cybersecurity, and mandatory reporting of serious safety incidents
- Legislative deadline
- OpenAI urged action before Congress adjourns in December
- State-level shift
- OpenAI endorsed four California bills, including some it had previously opposed
- Recent model release
- OpenAI released GPT-6 Astra one week before the policy statement
- Open-model position
- OpenAI said a national safety framework should not restrict publicly released open models