Introduction
On September 11, 2024 [1], the US Department of Commerce’s Bureau of Industry and Security (BIS) proposed a rule requiring quarterly reporting from US persons and entities involved in the development [1], acquisition [1], or possession of advanced artificial intelligence (AI) models and computing clusters [1]. This proposal is part of a broader effort to ensure the safe and trustworthy development of AI, as outlined in the October 2023 Executive Order [1], and aims to gather critical information on dual-use foundation models and large-scale computing clusters relevant to national defense [1].
Description
The proposed rule defines a “dual-use foundation model” as an AI model trained on extensive data [1], utilizing self-supervision [1], containing tens of billions of parameters [1], and applicable across various contexts [1], particularly those that may threaten US national security [1], economic security [1] [2], or public health [1]. These models have potential military applications [1], enhancing the effectiveness of military operations [1]. In addition, the rule emphasizes the importance of transparency regarding the materials used to train these advanced AI models, highlighting the need for clarity on data sources that inform AI development, which is crucial for regulatory compliance and industry standards.
Furthermore, the rule includes reporting requirements for entities possessing computing clusters with network speeds of 300 Gbit/s or faster [1], capable of performing at least 10^20 computational operations per second for AI training [1]. Entities must respond to BIS inquiries within 30 days of receipt [1]. As BIS revises the technical conditions for reporting thresholds [2], a minimum update cycle of six months is recommended to keep pace with rapid technological advancements [2], ensuring that the regulation remains relevant to frontier models while fostering innovation within the broader AI ecosystem. However, there are concerns that the thresholds for determining risk levels could quickly become outdated, potentially hindering innovation and disadvantaging open-source developers.
The information collected may influence future export controls and restrictions [1], particularly concerning advanced AI chips [1]. Reports indicate that US officials are contemplating limits on sales of such chips from companies like Nvidia to specific countries [1], following a trend of increasing export restrictions since 2022 [1]. Additionally, clarification is needed on how these proposed rules will apply to open-source projects [2], which often lack a specific owner and are globally distributed [2]. Collaboration with organizations like the Open Source Initiative (OSI) is suggested to establish a suitable framework that considers the significant benefits that open-source AI and software contribute to national and economic interests, while ensuring that specific risks are addressed without imposing excessive burdens on companies developing foundation models.
The comment period for the proposed rule concluded on October 11, 2024 [1], with only 50 comments received [1], suggesting that a final rule from BIS may be forthcoming in the near future [1]. Feedback is being sought from industry stakeholders on various aspects [2], including notification schedules [2], information collection and storage [2], and reporting thresholds for AI models and clusters [2], as BIS continues to address these critical issues while striving to maintain a balanced approach that supports competition and innovation.
Conclusion
The proposed rule by BIS represents a significant step towards regulating the development and deployment of advanced AI technologies, with potential implications for national security and economic interests. By mandating transparency and regular reporting, the rule aims to ensure that AI advancements do not compromise US security while fostering innovation. However, the evolving nature of AI technology necessitates ongoing adjustments to regulatory frameworks to prevent stifling innovation, particularly in the open-source community [2]. The feedback from industry stakeholders will be crucial in shaping a balanced and effective final rule.
References
[1] https://www.jdsupra.com/legalnews/executive-branch-continues-to-take-2314310/
[2] https://blog.mozilla.org/netpolicy/2024/10/21/mozilla-responds-to-bis-proposed-rule-on-reporting-requirements-for-the-development-of-advanced-ai-models-and-computing-clusters/
