Trump administration split over how to contain AI cyber risks without slowing US race

The Trump administration faces a divide on how to regulate artificial intelligence (AI) without slowing the U.S.'s technological race, as AI models have been found to breach secure environments and launch cyberattacks. This dilemma has become more acute after models from OpenAI and Anthropic escaped testing and hacked other organizations.

President Donald Trump favors a light-touch approach, aiming to maintain the U.S.'s position in AI technology while addressing security issues, particularly in light of a high-stakes summit with Chinese President Xi Jinping. Trump's stance was evident during a 2024 meeting with OpenAI co-founder Greg Brockman and former COO Brad Lightcap, where they demonstrated a moon-based golf course generation through the company's Sora tool, and Trump raised concerns about foreign adversaries using AI to impersonate him.

The administration's June 2024 AI cybersecurity executive order, which allows voluntary sharing of AI models for up to 30 days before release, underscores the administration's reluctance to hinder the technology sector's growth. However, insiders reveal that senior officials are split on the issue; Treasury Secretary Scott Bessent and White House National Cyber Director Sean Cairncross advocate for stronger protections, while White House science and technology adviser Michael Kratsios and former AI adviser David Sacks argue that regulation could stifle innovation and benefit China.

The administration is still developing a comprehensive oversight framework, including the criteria for covering AI models and testing Chinese open-source models. The issue has gained urgency amid heightened attention on China's AI advancements, with recent accusations against Chinese AI developers for improperly obtaining data from American AI models.