Trump Administration Controls Access to Frontier AI Models

The Trump administration is moving to exert greater control over the release of advanced artificial intelligence models, with sources indicating a new directive that will empower the government to decide which companies and entities gain access to the latest frontier AI systems. This marks a significant shift from the previous practice where leading AI developers unilaterally determined access.

This policy adjustment comes at a critical juncture for the AI sector, which faces escalating national security concerns and intensifying global competition. Companies like Anthropic and OpenAI, which have been at the forefront of developing powerful AI models, have historically managed their own release protocols. Anthropic, for instance, utilized its Project Glasswing initiative to grant select partners access to its cutting-edge Mythos cybersecurity model. Similarly, OpenAI has managed its GPT-5.6 and cybersecurity model releases through a consortium named Daybreak.

However, recent events suggest a more direct government intervention. Reports indicate that the administration previously blocked the release of Anthropic’s Claude Mythos 5 and Fable 5 models, citing national security concerns. Access was only reinstated after extensive negotiations. In another instance, OpenAI was reportedly asked by the administration to restrict access to its GPT-5.6 release to “trusted partners.”

While a White House official stated that the administration does not approve AI releases from private companies and that all government engagements are voluntary, emphasizing that companies retain full control over release timing and scope, the sequence of events points to a more assertive government posture. The official reiterated the administration’s commitment to collaborating with AI labs to enhance security without hindering innovation.

This evolving landscape presents a complex challenge for the U.S. AI industry. The administration is attempting to balance the imperative to foster innovation with the urgent need to mitigate the risks associated with advanced AI, particularly in the face of growing cybersecurity threats and the rapid advancement of AI capabilities from international competitors.

The emergence of sophisticated AI models from China, such as Moonshot AI’s Kimi K3, which reportedly rivals or even surpasses U.S. frontier models in certain benchmarks, underscores the urgency of the situation. Industry observers have expressed concern over losing the AI race if development is unduly hampered by regulatory hurdles or if U.S. companies are outpaced due to slower innovation cycles.

In line with this new direction, the administration has launched “Gold Eagle,” a program designed to foster public-private collaboration in identifying and addressing cyber vulnerabilities in AI systems. A source familiar with the initiative suggests that this “clearinghouse” mechanism will involve the White House in approving access to new AI models.

This strategic pivot by the administration raises questions about the future of existing company-led access programs like Project Glasswing and OpenAI’s Daybreak. Moving forward, it appears that such releases may necessitate explicit government authorization for partner inclusion, signaling a more centralized approach to managing the rollout of critical AI technologies. The administration’s actions reflect a proactive stance to ensure that the development and deployment of AI align with national security interests while aiming to maintain American leadership in this transformative field.