At the G7 Summit on Wednesday, world leaders like French President Emmanuel Macron and Indian Prime Minister Narendra Modi voiced concerns that the U.S. could cut off their countries’ access to top American AI models at any time. The meeting, held in a high-stakes atmosphere, brought together not only heads of state but also top executives from leading AI companies, including Anthropic CEO Dario Amodei, OpenAI CEO Sam Altman, and U.S. President Donald Trump.
During a working lunch, Macron warned that if the U.S. “from one day to the next can turn off the switch,” it could not only harm the economies of European customers but also damage the AI firms themselves. His remarks came just days after the Trump administration blocked Anthropic from exporting two of its newest models—Mythos 5 and Fable 5—on national security grounds. The order was triggered by Amazon flagging to the White House that certain safety guardrails in the models could be bypassed, although cybersecurity experts have argued that similar capabilities exist in other freely available models, including those from OpenAI.
The incident has exposed a profound risk that many international companies have been grappling with: any organization that builds on U.S. AI infrastructure now has to reckon with the possibility that access can be revoked overnight, for reasons they may never be told. This uncertainty undermines the reliability of American AI as a foundation for global innovation.
Prime Minister Modi expressed his own concerns, according to reporting from the Financial Times, stating that democratic nations must have unfettered access to top AI models to protect critical infrastructure. He emphasized that restrictions like the one on Anthropic set a dangerous precedent, especially as countries like China continue to advance their own AI capabilities.
Aidan Gomez, co-founder and CEO of Canadian enterprise AI firm Cohere, released a statement echoing these fears: “The recent restriction on access to Anthropic’s models confirms what we at Cohere have known all along: that companies and democratic nations remaining dependent on a small handful of big tech companies is dangerous to resilience. Digital sovereignty is not just about market competition or any one company or nation. It’s about who controls the foundational technology that will shape our economic security and national sovereignty for decades to come.”
During the G7 meeting, leaders discussed the creation of a “trusted partners” scheme that would grant non-U.S. nations access to advanced AI models from firms like Anthropic and OpenAI. The goal is to maintain an open trade network that bypasses U.S. restrictions. Both countries and companies could be designated as trusted partners, provided they use the models to develop stronger defenses against rivals like China. However, it remains unclear how far this partnership would extend or whether it can protect a startup in Paris or Bangalore whose product suddenly breaks due to a cutoff.
Macron noted that it would make sense for Washington to back such a scheme and ensure that access to models like Mythos is granted more broadly. He argued that nobody would want to buy U.S. AI access if it could disappear overnight. The comments underscore the delicate balance between national security and economic cooperation.
The broader context of these discussions is the global push for AI sovereignty. Europe, India, and other regions are investing heavily in domestic AI capabilities, but American models continue to pull ahead in performance and capabilities. The case for sovereignty becomes more difficult when the most advanced tools are controlled by a single nation, especially one that can revoke access arbitrarily.
This situation is reminiscent of earlier geopolitical tussles over technology, such as semiconductor export controls imposed on China. However, AI presents a unique challenge because the models themselves are inherently dual-use—they can be applied to both civilian and military purposes. The United States has justified its restrictions on the grounds of preventing adversaries from exploiting advanced AI for malicious ends, but allies are now questioning whether they are collateral damage in that strategy.
Anthropic’s Fable 5 and Mythos 5 models were designed to push the boundaries of reasoning and safety, but the export ban has stalled their global deployment. The company has not disclosed the specific vulnerabilities that Amazon reported, but industry insiders speculate that the guardrails could be bypassed through adversarial prompts or fine-tuning. Critics argue that such capabilities are not unique to Anthropic’s models and that banning them alone does little to address the broader risk.
The G7 Summit also touched on the need for a multilateral framework for AI governance. Leaders agreed that transparency and safety standards should be harmonized across nations, but deep disagreements remain about the role of government oversight versus industry self-regulation. The Trump administration has favored a light-touch approach, while European leaders have pushed for stricter laws like the EU AI Act.
For startups and enterprises that rely on American AI, the uncertainty created by the Anthropic ban is already having ripple effects. Companies are scrambling to diversify their AI suppliers, looking to open-source models and regional providers. However, most open-source alternatives still lag behind the capabilities of frontier models from Anthropic and OpenAI. The gap is narrowing, but not fast enough for many mission-critical applications.
In India, Prime Minister Modi has championed the “India AI” initiative, which aims to build a sovereign AI ecosystem. The country has invested in indigenous large language models and data centers, but the need for advanced American models remains acute for sectors like healthcare, finance, and defense. Modi’s comments at the G7 reflect a growing frustration among emerging economies that are caught in the crossfire of U.S.-China tech competition.
European leaders are similarly concerned. The European Union has poured billions into projects like the “EuroHPC” supercomputers and the “Trustworthy AI” label, but these efforts cannot match the rapid iteration cycle of Silicon Valley. Macron has proposed a “European AI Fund” to accelerate homegrown models, but he acknowledges that collaboration with U.S. firms is unavoidable in the short term.
The concept of a trusted partners scheme is not new—similar arrangements exist in other sensitive technology domains, such as nuclear energy and cybersecurity. However, applying it to AI is fraught with political and technical challenges. How do you verify that a partner is using AI models only for defensive purposes? What happens if a trusted partner unintentionally leaks the model or if its guardrails are compromised? These questions remain unanswered.
Moreover, the scheme could create a two-tier system: approved nations with unfettered access versus those left behind. This might exacerbate the digital divide and push excluded countries closer to China, which offers AI without the same conditions. Indeed, Chinese AI firms have been aggressively courting partners in Africa, Southeast Asia, and Latin America, promising open access—though with a different price tag in terms of data privacy and political influence.
The Trump administration has not publicly commented on the trusted partners proposal, but officials have signaled a willingness to consider it as a way to maintain U.S. influence in the global AI supply chain. However, internal divisions remain between security hawks who want to hoard cutting-edge AI and economic pragmatists who worry about losing markets to China.
Meanwhile, Anthropic itself is in a difficult position. The company has positioned itself as a safety-first AI lab, but the export ban threatens its international revenue streams and its reputation as a reliable partner. Dario Amodei, speaking at the G7, reiterated the company’s commitment to safety but urged governments to work together on shared standards rather than unilateral bans.
OpenAI CEO Sam Altman, who also attended the G7 lunch, has been more vocal about the risks of overregulation. He warned that too many restrictions could push AI development underground or to countries with fewer safeguards. Altman has proposed a global AI licensing framework modeled on nuclear regulatory agencies, but that idea has gained little traction so far.
The G7 discussions are part of a broader global conversation about AI governance. In May 2026, the United Nations will host an AI summit in New York, where many of these issues will be debated. The outcome of that summit could shape the rules of the road for the next decade. For now, the only certainty is uncertainty: the world wants American AI, but it wants it on terms that ensure reliability and sovereignty. Until a workable framework is found, the threat of the switch being turned off will hang over every international AI partnership.
Source: TechCrunch News