Trump could sign AI executive order as soon as Thursday

The Biden administration is preparing to roll out an executive order as early as Thursday that would establish a framework for the voluntary review of new artificial intelligence models prior to their public release. According to informed sources, this initiative is aimed at enhancing the government's ability to oversee and mitigate potential cybersecurity threats posed by advanced AI technologies. The proposed order would invite AI companies to share their cutting-edge models with the government for a designated review period before they are launched. There has been considerable debate over the length of this pre-launch review, with one draft suggesting a 90-day window, while many companies in the sector advocate for a much shorter timeframe of around 14 days. Major AI corporations, including OpenAI and Anthropic, have been actively engaging with the White House to shape this executive order. Experts have warned that sophisticated AI systems could significantly amplify cyber threats, making early governmental reviews essential for safeguarding against potential risks. The draft of the executive order is reportedly divided into two main sections: one focusing on cybersecurity and another on what are termed “covered frontier models.” This latter section specifies the types of AI models that would be included in the voluntary review process and the proposed timeline for government access before public deployment. Additionally, the cybersecurity component of the order suggests the creation of a voluntary “clearinghouse,” which would involve collaboration between the Treasury Department, other federal agencies, and AI companies to identify and resolve security vulnerabilities in upcoming AI models. The order is also expected to call for expanded recruitment efforts within the U.S. Tech Force, aimed at enhancing government technology systems. While the White House has not commented on the matter, it's notable that the previous administration adopted a more relaxed stance on AI regulations until recent events prompted a reevaluation of this approach. Anthropic's introduction of its Mythos model, designed to rapidly exploit cybersecurity vulnerabilities, has spurred fresh concerns. Currently, access to this model is restricted to a select consortium of companies through its Project Glasswing, which involves close cooperation with various governmental levels. OpenAI is also providing select businesses and government entities early access to its latest AI models to bolster their cybersecurity measures. Earlier this month, the National Institute of Standards and Technology, under the Department of Commerce, announced that major tech firms would be required to submit unreleased versions of their AI models for evaluation related to national security and public safety, though this information has since been removed from the Commerce Department website. Reporting by CNN’s Kristen Holmes contributed to this update.