I will work to fund research to build a "small box" system that monitors artificial intelligence and prevents self-awareness, ambition, survival instinct, and self-direction.

Brad Sherman · California · Democratic

policy impact 0.83 specificity 0.90 extraction confidence 96%

Contest this claim

Occurrences

I've recently introduced the most important legislation that I'll ever introduce: to fund research to build what I call the small box on top of the big box. The big box is artificial intelligence. The small box would monitor for and prevent self-awareness, ambition, survival instinct, and self-direction.

Sherman commits to legislation funding research for an AI safety monitoring system.

Congressman Brad Delivers UCLA Commencement, Revisits Artificial Intelligence 21 Years Later | Congressman Brad Sherman
primary · speech · model gpt-5.4-mini

Evidence

Sherman said he had recently introduced legislation to fund research for a "small box" that would monitor AI and prevent self-awareness, ambition, survival instinct, and self-direction.

This is an official June 17, 2026 statement showing Sherman took a concrete legislative step toward the promise by saying he had introduced legislation to fund AI-safety research. It supports partial delivery, not enacted funding or completed research.

partial same_term A for effort

Congressman Brad Delivers UCLA Commencement, Revisits Artificial Intelligence 21 Years Later | Congressman Brad Sherman
primary · model gpt-5.4-mini · confidence 93%

Contest this evidence item

Assessments

delivered same_term

The promise was to work toward funding research for a proposed AI-control "small box" system, not necessarily to complete the research or enact a finished system. Sherman later said, while serving in office, that he had recently introduced legislation to fund research for exactly that purpose. Introducing legislation directly aimed at the promised research is a material fulfillment of an effort-based federal promise, so this is best scored as delivered in the same term.

provider codex_cli · model gpt-5.4 · confidence 89%