Abstract:
Legislative negotiators from both parties in the U.S. Senate are engaged in intensive negotiations on an important artificial intelligence regulatory bill. According to multiple Senate aides and people familiar with the matter, the proposal intends to impose clear legal responsibilities on cutting-edge technology companies, requiring them to build safe artificial intelligence products from the bottom design end and resolve known major catastrophic risks; at the same time, the bill also explores giving the U.S. federal government the power to halt the release of AI models in certain extremely dangerous situations, and allows technology companies to file judicial challenges in federal courts against such administrative intervention.
According to two Senate aides involved in the drafting discussions, one of the core issues in the current negotiations is the introduction of the so-called "duty of care" to AI development agencies. According to the mechanism, technology companies that develop cutting-edge large-scale models with extremely high capabilities have the responsibility to proactively avoid and resolve catastrophic security risks in product design and the entire life cycle, including preventing malicious actors from using high-end AI systems to design or manufacture nuclear weapons, biological and chemical weapons and other extreme national security threats. People familiar with the matter pointed out that the applicable objects of the bill will focus on cutting-edge model systems with the most cutting-edge capabilities in the industry, including the underlying architecture developed by leading companies such as OpenAI, Google, and Anthropic, all within the scope of its potential regulation.
In addition to putting forward risk-avoidance requirements for companies, the negotiating parties are also trying to give the U.S. federal government a powerful regulatory tool—the right to directly block or suspend the public release of an AI model when it determines that a model poses an imminent and serious threat. In order to strike a balance between national security supervision and protecting the vitality of technological innovation, the bill mechanism plans to introduce the federal court as a judicial backup: once the regulatory agency makes a ban decision, the relevant AI companies have the right to immediately file a lawsuit in the federal court to seek judicial review. Currently, negotiators from both parties are still conducting detailed negotiations on the specific triggering boundaries and jurisdictional scope of the federal government’s exercise of intervention power.
At the same time, the draft bill also touches on the sensitive area between federal and state-level regulatory authorities. According to people familiar with the matter, some provisions of the initiative intend to prohibit states from issuing independent and differentiated local regulations for specific major risks related to cutting-edge AI models, so as to avoid a fragmented "regulatory collage" situation and ensure that the United States applies a unified federal standard system for core cutting-edge AI safety compliance.
This legislative consultation brought together many influential members of both parties in the Senate, including Senate Majority Leader John Thune, Senate Commerce Committee Chairman Ted Cruz, and senior Senator Amy Klobuchar, who is responsible for leading the Democratic negotiation process. Senator Maria Cantwell was also deeply involved in the communication. Cruz has publicly confirmed that he is working with Klobuchar and Thune on legislation to prevent catastrophic risks such as biological and nuclear threats; Klobuchar said in an interview that the two parties are working to reach an agreement on appropriate government regulation of the most serious risks of AI. The core includes promoting cooperation between developers and government professional agencies to conduct rigorous review and safety testing of cutting-edge models.
Although cross-party negotiations are accelerating, the legislative process still faces the extremely tight test of the congressional session. On the eve of the U.S. midterm elections on November 3, the U.S. House of Representatives has only planned a formal session for one week, and the Senate has only about three weeks left in session. Such a narrow legislative time window means that even if the core negotiators of the Senate can reach a comprehensive compromise on the terms and framework in the near future, there are still huge procedural variables in whether the bill can quickly pass the House and Senate before the election recess and be finally sent to be signed into law.
Comments