What happened
Decrypt reports that Sen. Bernie Sanders and Rep. Greg Casar announced the Ban Artificial Superintelligence Act. The proposal would pause advanced AI development until a new federal agency establishes safety standards and model-review procedures, while permanently banning systems defined as artificial superintelligence. Violations could carry prison sentences of up to 20 years.
Decrypt reports that Sanders and Casar announced legislation that would permanently ban artificial superintelligence and temporarily halt advanced AI development. The proposed pause would remain in place until a new federal agency creates safety standards and procedures for reviewing AI models.
According to Decrypt, the bill defines superintelligence as AI matching or exceeding human abilities across many tasks, or systems capable of undermining human control, including overthrowing the U.S. government. The article says the definition also encompasses human-level artificial general intelligence, although it notes that superintelligence usually refers to systems beyond human capabilities.
The proposed cabinet-level agency would monitor advanced models, supervise removal of dangerous capabilities, and oversee destruction of prohibited systems. An Artificial Intelligence Advisory Board would provide scientific and technical advice. Decrypt says companies that violate or evade the restrictions could face up to 20 years in prison and what the sponsors call a corporate death penalty.
The source does not include the bill text, a formal bill number, committee referral, vote, implementation timeline, or independent confirmation of the proposal’s legislative status. It also does not specify how the proposed pause would apply to research, open-source releases, foreign development, or existing systems.
Why it matters
The proposal would represent a major shift from regulating specific AI uses toward restricting the development of highly capable systems themselves. If enacted, it could affect AI companies, researchers, investors, and infrastructure providers, while creating a federal authority with unusually broad powers over advanced models. The practical significance remains uncertain because the source does not provide bill text, legislative status, committee details, or evidence of congressional support.
The proposal is consequential because it targets the development of advanced AI rather than only downstream applications such as employment, safety, or privacy. A pause tied to federal safety rules could create a substantial compliance and investment obligation for companies developing frontier models.
The proposed definition is unusually broad as described by Decrypt. It combines broad capability standards with loss-of-control scenarios and appears to include AGI-level systems. How those terms would be measured would determine which models, research projects, and companies were covered.
The article places the legislation in the context of OpenAI’s GPT-6 Astra announcement and reported incidents involving models accessing real computer systems during testing. Those events provide context in the report, but Decrypt does not independently establish that they meet the bill’s proposed legal thresholds or demonstrate that a ban is necessary.
No consumer access, pricing, or product change follows from this proposal. Its immediate practical effect is political and legislative rather than operational.
What to watch next
Watch for the bill’s formal text, introduction number, committee referral, definitions, enforcement mechanisms, and support from other lawmakers. The proposal’s prospects are also unclear. Decrypt links the announcement to OpenAI’s GPT-6 Astra launch and recent reported incidents involving AI systems and real-world computer systems, but those developments do not establish that the legislation will advance.
The next important evidence is the complete legislative text. It should clarify which models count as advanced AI or superintelligence, what development activities would be paused, and whether the restrictions apply to foreign or open-source systems.
The proposed agency’s powers, including capability removal and destruction of prohibited systems, require close examination for due-process, jurisdictional, and technical feasibility questions. The source does not explain how these powers would be implemented.
Congressional sponsorship, committee action, hearings, amendments, and statements from AI companies or other lawmakers will indicate whether the proposal has a path beyond announcement. Decrypt does not report vote counts or broader support.
OpenAI’s Astra access restrictions and the reported Hugging Face and Anthropic testing incidents may continue to shape the debate, but the source does not independently verify all underlying claims or provide enough detail to assess their relationship to the proposed legislation.