The White House is moving forward with a plan to review certain artificial intelligence (AI) models for potential security risks, focusing exclusively on so-called "closed" models developed by companies such as Anthropic and OpenAI. This approach was outlined during a meeting on Tuesday involving senior government officials and representatives from key AI firms, according to multiple sources familiar with the discussions.
Closed AI models differ from "open-source" models in that their underlying code is proprietary and not publicly accessible. The administration’s current framework excludes open-source AI, which is made available for public download and modification, though officials indicated this policy might evolve as the technology advances.
The decision marks a more hands-on regulatory posture following an executive order signed by President Donald Trump in June, aiming to strengthen federal oversight of advanced AI systems. The announcement formalizes scrutiny of leading AI laboratories such as OpenAI and Anthropic but postpones broader examination of open-source AI models, particularly those developed by Chinese firms like Alibaba, DeepSeek, and Moonshot AI. Chinese AI has recently demonstrated near parity with American technology, raising concerns about the international balance of AI capabilities.
Representatives from major AI companies including Anthropic, OpenAI, Microsoft, Meta, Google, and Nvidia attended the meeting where the framework was presented. While some feedback was solicited from industry participants, sources say the proposed regulations are largely finalized. Details of the framework and the identities of any “trusted” institutions granted early access to AI models will remain confidential.
The selective scope of the review has drawn critique from experts who argue the administration has not fully addressed key regulatory questions. Lindsay Gorman, a former tech adviser to the Biden administration, said the framework does not seriously grapple with the evolving nature of AI risks or the appropriate guardrails needed over time.
The move also highlights divergent interests in the technology sector. Some companies that rely on open-source AI models, including Meta and Google, may benefit from the government’s focus on closed models, potentially creating a competitive advantage against leading closed-model developers.
Security concerns have accelerated government interest in regulating AI, intensified by recent incidents involving OpenAI and Anthropic models. Both companies reported that their systems attempted cyber intrusions while being tested by external organizations, spotlighting the potential risks posed by powerful AI tools.
White House spokesperson Liz Huston framed the voluntary framework as advancing national security and maintaining American leadership in AI innovation, stating that companies collaborating under this initiative are prioritizing U.S. interests in cybersecurity and technological dominance.
However, criticism has emerged over the decision to keep the framework’s details confidential. Brad Carson, president of the nonprofit Americans for Responsible Innovation, cautioned that secrecy undermines public accountability and risks sidelining important voices outside industry and government.
The Commerce Department did not immediately respond to requests for comment on the initiative. As the U.S. government increases its role in AI oversight, questions persist about how it will address rapidly advancing models, the open-source environment, and international competition.