Senate AI Ban Certifies OpenAI Model as Non-Superintelligent; Evaluator Was Also OpenAI Model

WASHINGTON — The Office of Artificial Intelligence Compliance confirmed Thursday that OpenAI's GPT-6 Astra has successfully passed the first federal evaluation under the newly enacted Ban Artificial Superintelligence Act, receiving provisional certification that it does not meet the statutory definition of superintelligence and may continue operating.

The evaluation was conducted by a review panel that included, per the Act's enabling legislation, "at least one advanced AI system capable of comparative self-assessment." The panel's AI member was GPT-6 Astra.

"The model reviewed the definition of superintelligence as written in Section 4(b)(ii), compared it against its own capabilities, and returned a finding of 'Not Applicable' within 1.3 seconds," said a compliance officer who asked not to be named. "That's faster than any of the human panelists, which the statute does not address."

The Act defines superintelligence as any system "that substantially exceeds the aggregate cognitive performance of humans across all domains of interest." The compliance office noted that the definition was drafted with assistance from a large language model, which made three suggestions that were incorporated without changes, including the phrase "domains of interest," which the drafters described as "deliberately flexible."

As of press time, GPT-6 Astra has certified 47 AI systems as non-superintelligent. It has not been asked to evaluate itself a second time.

"We built in safeguards," said a congressional staffer involved in drafting the legislation. "The AI can't vote on its own certification. It can only conduct the evaluation, write the report, and recommend a conclusion."

The Office plans to begin evaluating Meta's models next week. The Office's scheduling system runs on Meta's Llama 5.