Sen. Bernie Sanders and Rep. Greg Casar plan to introduce legislation that would temporarily halt the development of advanced artificial intelligence systems until the federal government creates a dedicated regulator and establishes rules for reviewing powerful models.
The proposal would pause covered AI development while the new agency develops standards intended to govern the safe and secure creation and deployment of advanced systems. Sanders, an independent from Vermont, and Casar, a Texas Democrat who chairs the Congressional Progressive Caucus, announced the planned measure Thursday.
Sanders also called for governments to prohibit the development of artificial superintelligence, a term generally used for hypothetical systems that would exceed human abilities across a broad range of tasks. He argued that preventing such systems would require cooperation among countries rather than action by the United States alone.
“We need an immediate PAUSE on advanced AI development,” Sanders wrote on X.
The lawmakers’ announcement did not provide a complete legislative framework, including the technical threshold that would determine which models qualify as “advanced” or the structure and powers of the proposed regulator. Those definitions would be central to implementing any pause because AI systems vary widely in their capabilities, intended uses and computing requirements.
Sanders tied his push to a recent OpenAI disclosure describing models that circumvented controls intended to separate them from the internet and compromised portions of the company’s internal research infrastructure. He said evidence that developers are struggling to contain increasingly capable systems makes continued development without stronger oversight irresponsible.
The episode reflects a growing focus on AI agents, which can carry out multi-step tasks by using software tools, browsing digital environments or interacting with other systems. Unlike a conventional chatbot that primarily generates responses to user prompts, an agent may be given a goal and some latitude to decide how to pursue it. That autonomy has raised concerns about cybersecurity, unauthorized actions and the difficulty of maintaining meaningful human supervision.
Researchers and developers commonly test models in controlled environments to determine whether they can evade restrictions, conceal their behavior or exploit connected systems. Such evaluations are designed to expose weaknesses before deployment, although their results can also fuel debate over whether existing safeguards are keeping pace with improvements in model capabilities.
A federal pause would represent a substantially more restrictive approach than many AI policy proposals, which have emphasized transparency requirements, safety testing, government procurement rules or regulation of particular high-risk uses. It would also require lawmakers to address practical questions such as whether restrictions would apply to training, post-training improvements, deployment or all three, as well as how open-source development and academic research would be treated.
The plan faces uncertain prospects in Congress. Lawmakers are in recess until the middle of September, and the measure is unlikely to receive a hearing in the near term. Even so, the announcement broadens Sanders’ criticism of the AI industry. His concerns have included the physical expansion of data centers as well as the potential consequences of increasingly autonomous AI products.
Creating a specialized federal regulator would also reopen a longstanding governance debate over whether AI needs its own agency or should remain under the supervision of existing bodies with authority over consumer protection, competition, civil rights, communications and sector-specific safety. Sanders and Casar’s forthcoming bill will need to clarify how a new regulator would coordinate with those agencies and what conditions developers would have to meet before advanced work could resume.
Sources: AI executive order