Recently, the U.S. Department of Commerce announced a new proposal aimed at imposing reporting requirements on companies developing advanced artificial intelligence (AI) models and operators providing corresponding infrastructure.
This new regulation is in response to the Biden administration's executive order issued last fall regarding the "safe, reliable, and trustworthy development and use of artificial intelligence." Under this order, the Department of Commerce is required to establish and maintain permanent reporting standards.
Image source note: The image was generated by AI, provided by the image licensing service Midjourney
The core content of the new regulation is that any AI model requiring more than 1026 integer or floating-point operations for training must be reported. For models based on biological sequence data, the threshold for required operations is reduced to 1023.
Additionally, companies developing these models also need to disclose the capabilities of the models, information security measures, and the results of red team tests conducted on the models, to ensure that these AI technologies meet stringent safety standards, can withstand cyber-attacks, and reduce the risk of misuse by foreign adversaries or non-state actors.
The Department of Commerce's primary concern is that sufficiently advanced models could be used for cybercrime, and might even lower the threshold for developing biological, chemical, nuclear weapons, and dirty bombs. Therefore, it is essential to take action to detect, identify, and mitigate these threats. Infrastructure operators also need to report the network capabilities of their computing clusters, and if they exceed 300 Gbit/sec, or theoretical performance exceeds 1020 integer or floating-point operations per second. This increase in standards means that, compared to the past, the requirements for computing power are now stricter.
It is worth noting that these new regulations are not universally applicable to all businesses, and it is expected that relatively few infrastructure providers will meet the criteria. Commerce Secretary Gina Raimondo stated that with the rapid development of AI, this proposal will help maintain national security and defense.
Key Points:
1. 📊 The U.S. Department of Commerce proposes requiring large AI model developers and infrastructure operators to regularly report their technical capabilities and security measures.
2. 🔍 Projects training AI models requiring more than 1026 operations will be strictly monitored and required to disclose information security test results.
3. 🌍 The new regulations aim to enhance national security, preventing AI technology from being used for cybercrime or military purposes.