原文 · 未翻译
Trump's new executive order wants AI companies to voluntarily submit models for government safety reviews
Update – Jun 3, 2026
Added OpenAI's statement
The White House has issued a new executive order aimed at promoting AI innovation and security.
President Trump's order directs federal agencies to strengthen cyber defense of government systems within 30 days and expand AI-powered protection tools. The Department of Defense, the Cybersecurity and Infrastructure Security Agency (CISA), and the Treasury Department are tasked with setting up a clearinghouse for software vulnerabilities in partnership with the AI industry.
The order also creates a voluntary framework for so-called "covered frontier models." Under this framework, AI developers can submit their models to the government for safety testing before release. The order explicitly rules out any mandatory approval process. At the same time, it calls for ramping up criminal prosecution of AI misuse in cyber attacks and faster hiring of cybersecurity specialists.
The decision was likely shaped in part by the debate kicked off by Claude Mythos over the cybersecurity risks of AI. Google DeepMind, Microsoft, and xAI had already agreed with the US Center for AI Standards and Innovation (CAISI) to submit new AI models for review ahead of release. OpenAI and Anthropic entered a similar agreement back in 2024.
How voluntary is voluntary?
How willing this cooperation really is remains an open question. In the conflict between the Pentagon and Anthropic, the US government already showed it's ready to apply pressure to secure AI advantages, especially against China.
The new executive order makes the government's regulatory ambitions clear, even as it insists participation is voluntary and promises not to "stifle this innovation with overly burdensome regulation." Whether that balance holds up in practice is another matter. Anthropic, the same company that clashed with the Pentagon, called the executive order "an important step in strengthening America's leadership in AI."
OpenAI goes much further. In its strategy paper "Democratic Governance of Frontier AI," published today, the company calls for a binding national AI safety framework, mandatory pre-release testing of the most powerful models by CAISI, independent audits, and whistleblower protections.
AI safety decisions should be made by democratic governments, not individual companies, OpenAI argues. The executive order is "an important step forward," and the government should "now build on that foundation."
That OpenAI is pushing for stricter rules than the government itself likely reflects both genuine safety concerns and business interests. Mandatory pre-release reviews by a US agency would favor large, well-funded companies. Chinese open-source models like those from Deepseek or Qwen, which trail behind but not by much and cost far less, could face severe restrictions or be shut out of the US market entirely.