Bridging the Hill Gap in AI Safety Advocacy

Background and Executive Summary

With the fear of job displacement by Artificial Intelligence (AI) mounting, there has never been more public pressure on Congress to regulate the development and integration of AI. In response, frontier AI labs – companies that develop the most advanced AI – are joining hands to prevent regulations at both the state and federal levels. Therefore, AI safety advocates find the need to form a coalition to effectively coordinate and support measures against unchecked, unsafe AI developments. Hence, the AI Safety Alliance hosted a Happy Hour event for AI safety researchers and advocates.

The host, AI Safety Alliance, has been a rising figure in the AI regulation conversation across the aisle. John Doe, son of a prominent libertarian Senator Jack Doe, arrived early and conversed with the Alliance’s CEO, Brandon Smith. Other groups at the scene were BanAI and PolicyAI, while Hill staffers and unaffiliated individuals made up the rest. An advisor from FrontierAI RSVPed, but did not show up: fewer than 5 individuals representing the private sector.

Mr. Smith opened the event by highlighting the uniquely bipartisan nature of lobbying for AI-regulations. He also specially thanked libertarians for helping the Alliance. As the center of the coalition, he presented a balanced perspective on AI, with optimism on AI as a technology and a concern about the absence of safety measures.

After the opening remarks, attendees slowly split into two groups: policy experts and technical experts. The most active among the policy experts was BanAI, a London-based advocacy group that seeks to prevent the development of artificial superintelligence. Their director, Connor Leahy, and advisor Max Hernandez-Zapata conversed with most of the attendees. Three researchers represented IAPS, a technical research group on AI safety guidelines. IAPS and other technical researchers initially engaged in conversation with policymakers, but they visibly struggled to bridge technical findings with clear call-to-actions. Our organization has the opportunity to provide unique value in advancing AI regulations by bridging this gap.

Bipartisan Awareness and Opportunity for AI Regulation

A central discussion point at the event was on Senator Blackburn’s (R-TN) recent bill for a national AI framework. A section in the omnibus bill contains regulations on artificial superintelligence (ASI), defined with technical accuracy and enforced with a $1 million/day fine. The Alliance and its policy manager, Caleb Knapp, took credit for the bill’s ASI section. Mr. Steinhauser and ControlAI both stated that Senator Blackburn’s office has been receptive to threats of unsafe AI.

Meanwhile, legislative staffs from House Democrats shared similar stances. A legislative staff for Rep. Khanna (CA-17-R) shared that Democrats are pushing for regulations on AI, but many of them die at the committees. Nevertheless, both parties have demonstrated interest in regulating AI developments. The question lay in how.

Gaps Between Technical and Policy Experts

The gap between technical researchers and policy experts pose great challenge in drafting an effective regulation. Technical researchers have been publishing AI safety standards and guidelines, and IAPS has been the most active one. The IAPS researcher possessed an extensive background on potential AI regulations, but they were generally unable to communicate their recommendations in a concise pitch or an engaging story. Although Hill staffers and policy advisors for advocacy groups were attentive for the first hour, by the second hour, the groups at the event were visibly split between technical researchers and policy writers.

Gaps Within Policy Experts

There were gaps between advocacy groups as well, most notably between ControlAI and the Alliance. The Alliance for Secure AI took a moderate stance on AI, anticipating it as an inevitability that must be prepared for with appropriate regulations. ControlAI took a step further, contending that any artificial intelligence that exceeds human intelligence – that is, ASI – would present existential threats to humanity, and therefore must be wholly prevented. This attitude was most conspicuous in their reception to Senator Blackburn’s bill, to which both attendees from ControlAI showed dissatisfaction. The Alliance, being an author of a section of the bill, showed general satisfaction.

Recommendations Specific to Our Organization

In this context, our Organization should engage in two initiatives to accelerate collaboration for AI regulations. The advocacy space lacks a communicator who could translate technical concepts into intuitive ideas. Most of these individuals can be found in military spaces, where the practice of briefing complex technical and operational realities to policymakers is well established. Our Organization should actively recruit from this talent pool, and give them platforms to inform and educate strategies to bridge the gap. Second, our Organization is well-positioned to serve as a neutral convening body for the fragmented AI safety advocacy space. The tension between the Alliance for Secure AI's moderate regulatory stance and ControlAI's more absolutist position reflects a broader ideological divide that risks undermining legislative momentum. Without a coordinating actor, these groups will continue to present Congress with competing narratives, allowing frontier AI labs to exploit the disagreement. Our Organization should formalize a working group that brings these stakeholders together around shared near-term legislative priorities, while deferring more contentious questions about AGI and ASI to a longer-term policy roadmap.