US Body to Assess OpenAI and Anthropic Models Before Release
Two leading forces in artificial intelligence, OpenAI and Anthropic, have recently embarked on a groundbreaking partnership with a U.S. federal body. This collaboration aims to establish a precedent in AI safety by allowing early access to major AI models for comprehensive safety evaluations before their public release.
The deal has been sealed through a memorandum of understanding with the U.S. Artificial Intelligence Safety Institute, which falls under the umbrella of the Department of Commerce’s National Institute of Standards and Technology (NIST). This partnership not only paves the way for early model evaluations but also fosters a collaborative research environment dedicated to understanding and enhancing AI model safety and developing effective risk mitigation strategies.
“This agreement represents a significant milestone in our ongoing efforts to ensure the responsible development and deployment of AI technologies,” said Elizabeth Kelly, Director of the U.S. AI Safety Institute. Kelly emphasizes that safety is the cornerstone of technological innovation, underscoring the importance of these agreements in promoting a secure AI future.
The announcement arrives on the heels of a statement from OpenAI’s chief, Sam Altman, who took to the social media platform X to share news of this collaboration. Altman highlighted the potential of this partnership to advance the science behind AI evaluations significantly.
Since its establishment in February, following an executive order from the Biden administration, the AI Safety Institute has been at the forefront of efforts to create robust testing methodologies and research infrastructures for large language models. These efforts are not only theoretical but also aim to support practical, operational uses within the federal government.
With the new agreement, the Safety Institute will gain preemptive access to forthcoming models from OpenAI and Anthropic, allowing for an unprecedented level of scrutiny and feedback prior to and following their public releases. The partnerships underline a commitment to safety improvements and open the door for international cooperation with entities like the U.K.’s counterpart to harmonize AI safety standards globally.
This international collaboration reflects a wider ambition to address shared concerns about AI security, especially in light of increasing legislative attention at both federal and state levels in the U.S. There’s a collective recognition of the importance of establishing safeguards that protect against potential risks without hampering the innovative potential of AI technologies.
In support of national-level regulation, Altman argues that the U.S. must maintain its leadership position in AI development and policy. This stance is particularly relevant in the context of recent legislative efforts, such as California’s move to establish safety standards for advanced AI models—a bill met with mixed reactions from the AI community.
“Our partnership with the U.S. AI Safety Institute is a testament to our commitment to rigorous pre-deployment testing of our AI models,” stated Jack Clark, co-founder and head of policy at Anthropic. The collaboration aims to leverage the Safety Institute’s extensive expertise to enhance model safety proactively.
This initiative represents the first of its kind between the U.S. government and the tech industry, according to NIST’s announcement. It’s a notable acknowledgment of the proactive steps OpenAI and Anthropic are taking, alongside their involvement in the U.K.’s safety initiatives.
Furthermore, both companies are part of a group of 16 signatories that have pledged to develop and use AI responsibly. These commitments include investments in cybersecurity and efforts to combat deepfakes and other misleading AI-generated content through digital watermarking techniques.
As these partnerships begin to unfold, they mark a pivotal moment in the journey towards secure, responsible AI development. The collaboration between these AI powerhouses and the U.S. government could set a new standard for how AI safety is prioritized in the rapidly evolving tech landscape.
