Lawmaker urges Congress to ban AI self-improvement by 2028

The call for a ban

New York Assemblymember Alex Bores recently called on Congress to ban recursive self-improvement (RSI) as soon as possible this year. Bores, who identifies as a computer scientist and AI safety advocate, defined RSI as the practice of AI companies attempting to fully automate AI research. He specifically pointed to the goal of allowing AIs to build the next generation of AI systems.

The urgency of the request stems from the rapid progress currently seen in the industry. Bores noted that the timeline for these capabilities is approaching quickly, stating that the next US general election may occur before AI companies reach this specific threshold. He argued that the current technology is not ready for such a transition because any existing mistakes or misalignments could be multiplied with every new generation.

While Bores did not name a specific bill, his post gained significant attention on social media. He characterized the desire to limit RSI as a remarkably consensus opinion among various figures, including researchers and political figures. He suggested that while the exact mechanism of a ban can be debated, the ban does not necessarily have to be permanent.

Lab goals and timelines

The push for regulation follows specific public statements from the leading AI labs. OpenAI has publicly stated its internal goal of creating a true automated AI researcher by March 2028. They have already reported reaching a milestone in September 2026 to create an automated research intern, which currently performs 3.1 agent-workdays of effort for every workday of human labor.

Anthropic has also signaled that the path toward self-improvement is accelerating. The company's Institute noted that while recursive self-improvement is not inevitable, it could arrive sooner than most institutions are prepared to handle. They reported that their engineers are currently shipping 8x as much code per quarter than they did between 2021 and 2025.

Anthropic's chief scientist, Jared Kaplan, told the Guardian that a decision regarding letting AI recursively self-improve could come between 2027 and 2030. He described the possibility of AI self-improvement as one of the ultimate risks in the field. These timelines place the potential for significant shifts in AI development within the next few years.

Evidence of rapid progress

The concerns raised by Bores are supported by recent data regarding how these labs are scaling their operations. A report from METR, which was granted access to internal Anthropic data, suggests the company is likely improving model capabilities on well-specified tasks at about twice the rate they did between March 2024 and October 2025. This finding, while noted with high uncertainty, highlights the accelerating pace of R&D.

OpenAI's research organization is already integrating agents into its daily workflow to increase output. By using automated agents to handle research tasks, the lab is moving toward the goal of reducing the reliance on human labor for model development. This shift represents the transition from human-led research to the automated research capabilities that Bores is seeking to regulate.

The data shows a clear trend of increasing efficiency in both code production and research output. As these labs move toward their 2027 and 2028 targets, the technical infrastructure for self-improvement is becoming more robust. This creates a sense of urgency for policymakers who are watching the transition from manual oversight to automated iteration.

The political landscape

Bores's call to Congress is part of a broader trend of local and national efforts to regulate superintelligent AI. Just two days before Bores's post, the Bloomington, Indiana, Common Council voted to ban the distribution and deployment of superintelligent AI within the city. This local action mirrors the growing interest in setting boundaries on high-level AI capabilities.

In the federal sphere, other lawmakers have also called for restrictions on artificial superintelligence. For example, Representative Yassamin Ansari recently called for a ban on such systems. These efforts are often grouped into broader legislative discussions regarding the safety and governance of advanced AI models as they move toward autonomy.

The focus of these legislative efforts is shifting toward the internal processes of the labs. Rather than just looking at the products that customers buy, lawmakers like Bores are looking at the automation of research itself. This represents a move toward regulating the