Chuyện gì đã xảy ra
U.S. Rep. Josh Gottheimer (D-N.J.) announced two bipartisan bills on Friday: the American AI Security Act, which mandates a national security review for advanced AI models before release, and the China FIREWALL Act, which prohibits federal agencies from using Chinese-developed open- AI models. The bills are co-led by Republican Representatives Mike Lawler and Nick LaLota, respectively.
Rep. Josh Gottheimer announced the American AI Security Act and the China FIREWALL Act during an event at Fairleigh Dickinson University in Teaneck, New Jersey. The American AI Security Act, co-led by Rep. Mike Lawler (R-N.Y.), would require developers of certain advanced AI models to submit their technology to the National Security Agency for a mandatory pre-release review. The review would assess whether the model could be used to conduct serious cyberattacks or assist in developing chemical, biological, or radiological weapons. The proposal sets a 30-day deadline for this review, with one 30-day extension permitted, and allows for an expedited appeal process for companies.
The China FIREWALL Act, co-led by Rep. Nick LaLota (R-N.Y.), would prohibit the use of Chinese-developed open- AI models on government-issued devices and bar federal agencies from purchasing software that relies on these models. According to Gottheimer’s office, this bill expands upon existing restrictions on the DeepSeek model included in the fiscal 2026 National Defense Authorization Act by covering a broader range of Chinese-developed open-weight models. Gottheimer also announced plans to send a letter to National Cyber Director Sean Cairncross to seek information regarding the administration’s AI cybersecurity clearinghouse and its capacity to address autonomous AI-driven cyberattacks.
Gottheimer emphasized that there are currently no mandatory requirements for pre-deployment testing of frontier AI models, comparing the lack of regulation to standards in banking, food safety, and medicine. He stated that the goal is not to stifle innovation but to ensure safe deployment. He called on House leadership to bring Congress back before the election to consider these measures, noting that he has led more than a dozen bipartisan AI bills in the current Congress.
Chi tiết nguồn: wrnjradio.com ↗
Tại sao nó quan trọng
These proposals represent a significant legislative push to institutionalize and national security protocols. By mandating pre-deployment reviews by the NSA and restricting foreign-developed models in federal systems, the bills aim to address critical vulnerabilities in cyber and biological security. This marks a concrete step toward regulatory that go beyond voluntary industry standards, potentially setting a precedent for how the U.S. government manages frontier AI risks.
The introduction of these bills signals a shift toward mandatory, government-led oversight of frontier AI capabilities, specifically targeting dual-use risks in cyber and biological domains. The involvement of the NSA in pre-release reviews is a notable escalation from previous proposals that relied on voluntary industry self-regulation or broader executive orders.
The restriction on Chinese-developed open- models in federal systems reflects growing concerns about supply chain security and the potential for adversarial exploitation of AI infrastructure. By expanding beyond specific models like DeepSeek to a broader class of open-weight models, the legislation aims to create a more comprehensive barrier against foreign influence in critical government systems.
The bipartisan nature of the bills, with Republican co-sponsors, suggests a level of cross-party consensus on the need for AI security . This could facilitate faster legislative progress compared to more polarized policy areas, although the specific technical and operational details of the NSA review process remain a point of potential friction with the tech industry.
Cơ chế tương tác: Nó thực sự hoạt động như thế nào
Khám phá công nghệ cơ bản đằng sau sự phát triển này một cách tương tác.
crm_get_transaction(id='4092').Why can ethical evaluation not be reduced to one model score?
Xem gì tiếp theo
Monitor whether House leadership schedules these bills for a vote before the upcoming election, as Gottheimer has explicitly called for this. Also watch for reactions from AI developers regarding the feasibility of the 30-day NSA review timeline and the specific technical criteria that will define 'covered' models under the security act.
The immediate next step is whether House leadership agrees to schedule a vote on these bills before the election, as requested by Gottheimer. Without a floor vote, the bills may stall in committee or be delayed until the next congressional session.
Industry reactions to the 30-day NSA review timeline will be crucial. AI developers may argue that this timeframe is insufficient for thorough security evaluation, potentially leading to lobbying efforts to modify the review process or the definition of 'covered' models.
The scope of the China FIREWALL Act will be closely watched, particularly how 'open-' models are defined and whether the ban applies to models that are fine-tuned or hosted on third-party platforms. This could have significant implications for federal agencies that currently use open-source AI tools for various applications.