Biden Administration Calls for Increased Monitoring of Open-Weight AI Models

The Biden administration is calling for increased monitoring of open-weight AI models to assess potential risks and inform future regulations, while acknowledging the current lack of capacity to effectively respond to these risks.

Key focus areas for monitoring open-weight models: The National Telecommunications and Information Administration (NTIA) suggests focusing on three main areas to address the risks posed by open-weight AI models:

• Collecting evidence on the capabilities of the models to monitor specific risks
• Evaluating and comparing indicators of these risks
• Adopting policies that target the identified risks

Defining open-weight models and their unique challenges: Open-weight models are foundation models whose parameters are publicly released and can be downloaded by users, which may pose distinct opportunities and challenges in risk reduction compared to closed models.

• The NTIA acknowledges that both open and closed models have risks that need to be managed
• The agency emphasizes the importance of considering the marginal risk of open-weight models to avoid unduly strict restrictions compared to alternative systems with similar risk-benefit balances

Potential regulatory approach and industry reactions: The Biden administration’s focus on the risks of open models could indicate a regulatory approach similar to the European Union’s AI Act, which regulates AI based on the riskiness of use cases rather than the models themselves.

• Kevin Bankston, a senior advisor on AI governance, applauds the NTIA for not prematurely imposing new restrictions on the distribution of open foundation models due to insufficient evidence of novel risks
• AI model developers do not face immediate changes, as the NTIA is still in a fact-finding phase, but they may be subject to increased scrutiny when releasing model weights at their own discretion

Broader implications and remaining questions: The Biden administration’s proactive approach to creating guidelines around AI use, such as the AI Executive Order, demonstrates a commitment to addressing the potential risks associated with AI models. However, the lack of specific regulations at this stage leaves some uncertainty for AI model developers and companies working with foundation models.

• The flexibility and adaptability of federal agencies in response to the rapidly evolving AI sector will be crucial in effectively regulating and mitigating risks
• The potential impact of proposed state-level policies, such as California’s “Safe and Secure Innovation for Frontier Artificial Intelligence Models Act,” on AI development and smaller companies working with foundation models remains to be seen

As the NTIA continues its fact-finding mission and the AI landscape evolves, it will be essential to strike a balance between mitigating risks and fostering innovation in the development and application of open-weight AI models.