Ro Khanna Backed Legislation to Limit Recursive AI
The proposed bill aims to restrict advanced systems capable of recursive self-improvement to maintain human control.
Updated on Sept. 27, 2026 in Artificial Intelligence

Live Poll
Should the federal government ban the development of self-improving artificial intelligence systems?
Representative Ro Khanna has signaled support for legislation, introduced by Senator Bernie Sanders and Representative Greg Casar, that would ban recursive self-improving artificial intelligence. The policy reflects concerns among researchers that such advanced systems could eventually bypass human oversight.
Why it matters
This legislative push highlights growing tension between rapid advancements in AI capability and the regulatory frameworks needed to manage systemic risks. By seeking to limit recursive self-improvement, proponents aim to ensure that human control remains an integral part of high-level model development.
The proposed legislation targets systems designed for recursive self-improvement, a mechanism where an AI refines its own architecture or code without external guidance. The initiative seeks to establish enforceable standards, with Khanna proposing a working group of American and Chinese experts.
The players
Ro Khanna
United States Representative for a Silicon Valley district known for his legislative focus on technology policy and foreign affairs.
Bernie Sanders
United States Senator who sponsors legislation aiming to regulate advanced AI and limit system autonomy.
Greg Casar
United States Representative who co-introduced the bill to restrict advanced artificial intelligence development.
OpenAI
An AI research and deployment company that develops large language models and is now central to legislative discussions on AI limits.
Anthropic
An AI research laboratory focused on developing steerable, reliable AI systems and defining safety protocols for advanced models.
The details
Recursive self-improvement refers to autonomous agents that iteratively modify their own algorithms to enhance performance. These systems theoretically possess the ability to scale their intelligence exponentially, creating a scenario where the internal logic becomes opaque to human overseers. The proposed bill aims to preemptively ban these capabilities to prevent scenarios where control over the model is lost.
Timeline
The nuclear accord with Iran was signed in 2015.
Representative Ro Khanna appeared on Meet the Press on September 27, 2026.
The Tech Race
The push for an enforceable US-China pact marks a departure from purely domestic AI safety standards. It follows a precedent set by frameworks like the 2015 Iran nuclear accord, attempting to manage competitive tensions through multi-national technical agreements.
The proposed regulations currently target enterprise-level research and are not expected to restrict existing consumer AI tools. However, developers and industry stakeholders should monitor the upcoming working group meetings for updates on potential technical compliance requirements.
The takeaway
The move suggests a shift toward aggressive preemptive regulation of AI architectures that exceed human-directed learning. Watch for the findings of the proposed working group meetings, as these sessions will likely determine the technical feasibility of the suggested self-improvement bans.
What happens next
Representative Ro Khanna plans to meet with a group of 10 to 15 researchers from OpenAI and Anthropic to discuss the technical parameters of the proposed limits.
Further reading
For more on the current landscape of safety regulation, see the latest updates in Artificial Intelligence.
Source note: This article includes information reported by Tampa Free Press.
Live Poll
Should the federal government ban the development of self-improving artificial intelligence systems?









