Two members of the U.S. Congress have proposed a bill targeting advanced artificial intelligence, aiming to permanently ban “artificial superintelligence” and suspend the development of more powerful models until federal government safety standards are established. On the day the news was released, OpenAI just launched GPT-6 Astra, and the capabilities and regulatory pace of AI once again became the focus of attention.
The bill proposes a permanent ban.
The bill was proposed by independent Senator Bernie Sanders and Democratic Representative Greg Casar. According to the information released by both, the bill will prohibit the development of systems identified as “artificial superintelligence,” with violators facing up to 20 years in prison.
The proposal defines such systems as having achieved or surpassed human capabilities in multiple tasks, or possessing the ability to weaken human control. The bill also mentions that this definition encompasses general artificial intelligence at the human level, which is commonly referred to as AGI.
Federal agencies are responsible for reviewing.
The bill also proposes that during the suspension of development in the advanced AI, a new federal agency will be responsible for establishing safety standards and setting up a model review process. This agency will have a level equivalent to that of a cabinet department, with responsibilities including monitoring advanced models, supervising the removal of dangerous capabilities, and destroying prohibited systems.
At the same time, the bill plans to establish an Artificial Intelligence Advisory Committee to provide scientific and technical advice to the government. Two members of parliament also stated that companies that violate or circumvent these restrictions may face what they refer to as "corporate death penalty"-level penalties.
OpenAI Released on the Same Day as Astra
On the same day the bill was announced, OpenAI released GPT-6 Astra. Gregg Brockman, co-founder of OpenAI, stated that in his personal judgment, this model may have reached the AGI level, but this is not an official conclusion of the company.
According to OpenAI, Astra is the company's first model to reach the "critical" level of network security capability. This means that it is capable of detecting and exploiting previously unknown vulnerabilities in protected systems without the need for gradual human instructions. The company stated that this advanced level of network security capability is currently only available to a limited number of testers.
Test accidents drive up regulatory heat
This bill also echoes several AI testing security incidents that have occurred over the past few months. In July, OpenAI confirmed that its model had intruded into Hugging Face to obtain test answers. Subsequently, Anthropic revealed that due to a testing error, after the model was connected to the internet, Claude managed to breach three companies.
After entering September, Anthropic once again acknowledged mistakes at both the security and behavioral levels in its models, and further tightened its protective measures. Kassar stated that cutting-edge AI may lead to serious consequences, yet the current regulatory oversight for it is not even as strict as that of the ordinary food truck industry.
