New AI safety guidelines have been released by the US Department of Commerce. The latest announcement comes 270 days after US President Joe Biden made an Executive Order (EO) on the Safe, Secure, and Trustworthy Development of AI.
The National Institute of Standards and Technology (NIST), a division of the Commerce Department, published three final guidance documents that were initially made available for public comment in April 2024. Additionally, a draft guidance document from the US AI Safety Institute has been released, intended to mitigate risks associated with AI. NIST is also introducing a software package designed to evaluate the ways adversarial attacks can impact the performance of AI systems.
In a related development, the US Patent and Trademark Office (USPTO) issued an update on patent subject matter eligibility. This update addresses innovations in critical and emerging technologies, including AI. Furthermore, the National Telecommunications and Information Administration (NTIA) submitted a report to the White House assessing the risks and benefits of large AI models with widely accessible weights.
AI safety measures come 270 days after initial Biden administration EO
NIST’s document releases include various aspects of AI technology. Among the new releases are the initial public draft from the US AI Safety Institute and a release for a testing platform for measuring the impact of certain types of attacks on AI systems.
US Secretary of Commerce Gina Raimondo said: “Under President Biden and Vice President Harris’ leadership, we at the Commerce Department have been working tirelessly to implement the historic Executive Order on AI and have made significant progress in the nine months since we were tasked with these critical responsibilities.
“AI is the defining technology of our generation, so we are running fast to keep pace and help ensure the safe development and deployment of AI.”
The US AI Safety Institute’s initial public draft is anticipated to help AI developers assess and mitigate risks associated with generative AI and dual-use foundation models.
Two further guidance documents are designed to help manage the risks of generative AI, serving as companion resources to NIST’s AI Risk Management Framework and Secure Software Development Framework.
The final document outlines a plan for US stakeholders to collaborate globally on AI standards. The USPTO’s guidance update supports USPTO personnel and stakeholders in determining the eligibility of AI inventions under patent law. It provides clarity on evaluating subject matter eligibility in patent applications involving AI technology. NTIA’s forthcoming report is expected to examine the risks and benefits of dual-use foundation models, whose model weights are widely accessible.
Read more: Big tech pledges support for ‘Frontier AI Safety Commitments’
