U.S. Government to Vet New AI Models from OpenAI and Anthropic

The page discusses the recent agreements between OpenAI, Anthropic, and the U.S. government's AI Safety Institute to evaluate new AI models prior to their public release. It highlights the roles of the U.S. AI Safety Institute in assessing the capabilities and risks of these models, emphasizing the importance of responsible AI development. The content also touches on the implications for AI regulation, including California's recent legislative efforts and the collaboration with UK counterparts to enhance model safety. Additionally, it outlines the partnerships' goals of fostering innovation while addressing safety concerns in advanced AI systems.

Sep 2, 2024
According to reports from CNBC and NIST, OpenAI and Anthropic have agreed to share their new AI models with the U.S. government's AI Safety Institute for evaluation before public release, marking a significant step towards ensuring responsible AI development and deployment.

U.S. AI Safety Institute Role

notion image
The U.S. AI Safety Institute, established under the National Institute of Standards and Technology (NIST), will play a crucial role in evaluating and testing new AI models from OpenAI and Anthropic.
This initiative, announced in August 2024, aims to assess these models' capabilities and potential risks before and after their public release. 
The institute will provide feedback on safety improvements, working closely with its UK counterpart. Elizabeth Kelly, director of the U.S. AI Safety Institute emphasized that these agreements mark an important milestone in responsibly guiding AI's future, building upon NIST's 120-year legacy of advancing measurement science and technology standards.

OpenAI and Anthropic Partnership

notion image
Leading AI companies OpenAI and Anthropic have formalized agreements with the U.S. government to collaborate on AI safety research, testing, and evaluation.
These partnerships, announced in August 2024, allow the companies to share significant new AI models with the U.S. AI Safety Institute before and after public release. 
The collaboration aims to assess capabilities, identify potential risks, and develop mitigation strategies for advanced AI systems. 
Jason Kwon, OpenAI's chief strategy officer, expressed support for the institute's objectives and anticipated joint efforts to enhance safety practices and standards. 
Similarly, Jack Clark, Anthropic's co-founder, highlighted how the partnership leverages the institute's expertise to rigorously test models before widespread deployment, strengthening its ability to identify and address risks.

AI Regulation and Safety Implications

notion image
The agreements between OpenAI, Anthropic, and the U.S. government come at a critical juncture as lawmakers grapple with establishing appropriate regulatory frameworks for AI technology.
The California legislature recently advanced the Safe and Responsible Innovation for Artificial Intelligence Act (SB 1047), which would require AI companies to implement safety protocols before developing advanced foundational models. 
However, this bill has faced opposition from OpenAI and Anthropic, who argue that it could hinder innovation and negatively impact smaller open-source developers.
Concurrently, the White House is seeking voluntary pledges from major corporations regarding AI safety protocols, focusing on enhancing cybersecurity, researching discrimination issues, and developing watermarking for AI-generated content.

International Collaboration with UK

notion image
The U.S. AI Safety Institute plans to collaborate closely with its UK counterpart, sharing findings and feedback to enhance model safety. 
This international partnership reflects a growing recognition of AI development's global nature and the need for coordinated efforts to address its challenges.
By working together, the two institutes aim to provide comprehensive feedback to OpenAI and Anthropic on potential safety improvements to their models, fostering a more robust and globally aligned approach to AI safety.

References

The agreements between OpenAI, Anthropic, and the U.S. AI Safety Institute represent a significant step in collaborative AI safety research and evaluation.
These partnerships, formalized through memoranda of understanding, grant the institute access to major new AI models before and after their public release. 
The collaboration aims to assess capabilities, identify safety risks, and develop mitigation strategies for advanced AI systems. Elizabeth Kelly, director of the U.S. AI Safety Institute, emphasized that these agreements mark an important milestone in responsibly guiding AI's future. 
The institute plans to work closely with its UK counterpart to provide feedback on potential safety improvements, reflecting a growing recognition of the need for international cooperation in addressing AI challenges.
 

References