The Trump administration's tightening grip on artificial intelligence development is fragmenting global access to cutting-edge models, forcing companies and nations to choose between compliance and capability—a dynamic that threatens to concentrate technological power while leaving key allies and enterprises without reliable tools.
OpenAI announced three new models on Friday—GPT-5.6 Sol, Terra and Luna—but immediately limited their rollout to a small group of government-vetted partners, complying with a Trump administration request that came weeks after the same government forced rival Anthropic to disable access to two of its latest models entirely. The restricted access underscores a fundamental tension: while OpenAI said it "believes in broad access" and plans wider availability in coming weeks, the company also acknowledged that government vetting "keeps the best tools from users, developers, enterprises, cyber defenders and global partners who need them."
The administration's hands-on approach began earlier this month when President Donald Trump signed an AI executive order asking developers to voluntarily allow government assessment of model capabilities before full release. Though thin on specifics, the order has already reshaped the industry. Anthropic, which had been on a historic growth trajectory with run-rate revenue crossing $47 billion in May 2026, remains in active negotiations with Washington officials but has not disclosed when its banned models will return online.
The Global Vacuum Left Behind
The export restrictions have created immediate openings for competitors outside the U.S. regulatory framework. Two weeks ago, Chinese cybersecurity firm 360 unveiled Tulongfeng, a model positioned as competitive with Anthropic's offerings, which the Trump administration has already banned for non-Americans. Simultaneously, Tokyo-based startup Sakana AI launched Fugu, a frontier model explicitly marketed as delivering "frontier capability without the risk of export controls."
Sakana co-founder David Ha framed the moment as exposing a structural vulnerability: "Access to top models can disappear overnight," he wrote on social media, arguing that "relying on a single provider for national infrastructure is a risk the recent export controls made impossible to ignore." Ha and co-founder Ren Ito—both former Google researchers—are targeting Japanese businesses and government agencies looking to reduce exposure to U.S. export restrictions, with models optimized for Japanese language and culture.
Ren Ito acknowledged the delicate balance in his assessment: "U.S. models remain important to Asia," he said, while cautioning against what he called a "permanent realignment." Yet in an op-ed published last week in Project Syndicate, Ito urged the U.S. federal government to reconsider its approach, arguing that "AI should not become a technology that is hoarded; it should be one that is developed together." He specifically called for prioritizing access for America's closest allies.
The Concentration Risk
Ha's concept of "orchestration models"—systems designed to coordinate access across multiple AI providers—reflects a growing recognition that dependence on any single source of advanced technology carries unacceptable risk. "Collective intelligence is the practical hedge against this concentration of power," Ha wrote, positioning Fugu as a tool for distributing reliance rather than consolidating it.
OpenAI's own framing reveals internal tension about government oversight. The company stated, "We don't believe this kind of government access process should become the long-term default," while simultaneously working with the Trump administration to "develop a repeatable process for future model releases." This suggests OpenAI views current restrictions as temporary but accepts the principle of government assessment.
The models themselves reflect diverging technical priorities. OpenAI's Sol—described as its strongest offering yet—shows particular improvements in coding, biology, and cybersecurity, with the company noting it is "better at helping users fix vulnerabilities than it is at carrying out end-to-end attacks." The company emphasized that Sol "still does not cross into OpenAI's critical cybersecurity risk threshold, which is defined as bringing unprecedented new pathways to severe harm."
The Lasting Impact
What remains unclear is how permanent the shift toward localized AI alternatives will prove. Sakana has not proclaimed "a lasting shift away from U.S. AI in Asia," and the company's spokesperson characterized its model launch as "entirely coincidental" to the export ban—even as the company's marketing explicitly capitalizes on the moment. Yet the underlying dynamic suggests structural change: even if U.S. companies regain trust should export restrictions lift, local alternatives trained to understand local language and cultural nuance are already filling gaps and building customer relationships.
Anthropus's revenue dependence on Asian enterprise customers remains publicly unknown, but the weeks since the export order took effect have already produced viable alternatives from Tokyo and Beijing.
Why This Matters:
The fragmentation of AI access along national lines raises fundamental questions about how transformative technologies will be governed in a multipolar world. When the U.S. government restricts allies' access to advanced models—as it has done with Anthropic—it accelerates incentives for those allies to develop independent capabilities, potentially accelerating a technological divergence that could undermine both innovation and security cooperation. The concentration of frontier AI capability in a handful of U.S. companies, combined with export controls that prevent reliable access, creates pressure toward duplication of effort and reduced collaboration precisely when shared standards and transparent development might serve collective interests. Workers, enterprises, and governments in allied nations face immediate disruption to their technology roadmaps, while the long-term effect may be slower innovation across the field and reduced ability for democratic societies to collectively shape AI development toward shared values.