OpenAI reverses stance, calls for mandatory US AI safety rules
What's the story
OpenAI, the artificial intelligence (AI) giant, has reversed its earlier stance on regulatory measures. The company is now calling for mandatory capability-based national AI safety regulations in the US. This comes as a major shift from its previous opposition to stringent rules and warnings that early regulation could hand over leadership to China.
Regulatory framework
Chris Lehane outlines OpenAI's new position
Chris Lehane, OpenAI's Chief Global Affairs Officer, detailed the company's new position in a blog post on September 9.
He emphasized the need for a robust federal framework for frontier AI in the US.
This would include common testing, independent assessments, cybersecurity protections, and clear incident-reporting requirements aimed at "the 'handful of well-resourced laboratories' developing the most capable systems."
Legislative support
Four-part agenda proposed as new regulatory compact
Lehane's blog post also detailed a four-part agenda that looks like a blueprint for a new regulatory compact.
This includes supporting four California bills headed to Governor Gavin Newsom's desk.
The proposed legislation aims to establish a process for designating qualified, independent organizations to assess AI risks, create registration and accountability requirements for AI auditors among other things.
Safety endorsement
Industry-led monitoring standards for misalignment and serious incidents
Along with supporting California bills, OpenAI has also endorsed mandatory federal safety rules for frontier labs.
These include independent technical assessments, stronger cybersecurity measures, and incident reporting.
The company also backs industry-led monitoring standards for misalignment and serious incidents.
This was initially proposed as a voluntary effort but is now intended to inform federal policy.
Global coordination
International coordination on capability measurement, human control advocated
Lehane has also advocated for international coordination on capability measurement, human control, and when to slow or stop development "even if that means slowing the advancement of model capabilities."
This comes after OpenAI's chief scientist, Jakub Pachocki, warned that no lab has solved AI alignment at current scales and voluntary slowdowns may soon be necessary.
Rising fears
Policy shift follows alarming predictions about AI risks
The policy shift also comes after the resignation of Jacob Coxon, a former Anthropic and OpenAI researcher.
He left the industry entirely, saying leading labs are "racing straight to self-improving superintelligence" and that AI systems could be "out of control" by the end of 2027.
The launch of GPT Astra, which NVIDIA CEO Jensen Huang called Artificial General Intelligence (AGI), has also added to these concerns.