Two world super powers, the United States and the People’s Republic of China, are responsible for the bulk of the world’s foundational artificial intelligence (AI) development. Given their internal policies, it’s fair to address both as AI-first or AI-native countries. This places the rest of the globe in the position to innovate against the tide or become downstream “customers” providing data and training resources to foreign corporations.
The world’s most powerful and popular AI models aren’t ubiquitously available, however, even among users in countries where they’re made. Anthropic’s newest models, Claude Fable 5 and Claude Mythos 5, for example, are currently being withheld from users. The U.S. government, citing national security, imposed export controls on both models for all foreign nationals in the country and users outside, including company employees.
In a statement published on its website on 12 June, Anthropic agreed to follow the government’s order, but objected to its premise:
”We are complying with the government’s legal directive and are removing access to Fable 5 and Mythos 5 for all users. However, we disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people. If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers.”
The rapidly unfolding Mythos debacle demonstrates that the most powerful AI models will be tightly controlled and meted out only to those deemed worthy by the governments in control of the infrastructure they’re built on.
As developers dial in stronger models and, potentially, unlock artificial general intelligence (AGI), the rights of those who live or work in areas where access to these models is not guaranteed could become further eroded by the lack of technological resources.
A vehicle for global dominance
Assuming AGI is reached, there currently exists no agreed-upon framework for its distribution or qualifying access. While the US is currently considered to have an edge in the so-called AI race, there are no guarantees that its developers will be the first to produce a system widely recognized as being AGI-capable.
Many of the potential scenarios raise the question of how AGI rights become even more challenging in places where the absence of legal grounds can cause serious damage to society and governments.
“When the US government writes AGI requirements into federal contracts, that becomes global law overnight,” said Bryan Lopez, a senior director at Microsoft’s Federal Strategic Science Missions, in commentary to AGI Ethics News. “The same way FedRAMP quietly became the world’s cloud security baseline.” FedRAMP, the Federal Risk and Authorization Management Program (FedRAMP), enables U.S. federal agencies to assess, authorize, and monitor global cloud products and services.
This one-sided, US-based AI governance and rights issue is a discussion point at many intergovernmental summits and meetings. During its 17th Summit, BRICS, the intergovernmental organization, signed a Leader’s Declaration on Global Governance of Artificial Intelligence.
Featured interview
David Gunkel on AI Ethics and the “Relational Turn”
Read AGI Ethics News’ exclusive four-part interview with philosopher David Gunkel on moral standing, machine ethics, and the “relational turn” in AI.
“The relational turn doesn’t give us a different morality, it just says, here’s another way of looking at it.”
The declaration codified AI access as a matter of “rights” for countries of both developed and developing stature, rather than a manipulation tool for a few countries and the rich. In the context of futuristic AGI, ongoing research from around the globe highlights similar concerns.
The Spark Hunter Survey from AGI Ethics News, for example, showed that 54.8% of journalists from eight countries were “extremely concerned” about AGI development by a small number of powerful corporations or perhaps nations.
According to a policy report from the RAND Corporation, AGI development is divided into two categories: centralized and decentralized. Centralized AGI development happens when new or existing AI companies such as Anthropic, OpenAI, Google, Meta and Google make models to attain AGI capabilities. The US, in such a case, sets all the terms and conditions for the rest of the world to accept. Under this paradigm, US allies are typically given preferred access.
Decentralized AGI development, however, is where all countries contribute to a rising tide of development in hopes of creating an AGI system. These efforts are typically considered more brittle than centralized efforts as coordination can break due to any number of shifting technological, political, or ideological differences between participating nations and organizations.
Alex Levkin, patent attorney and founder of IPNOTE, shared his thoughts on distributed AI in an email with AGI Ethics News. “If compute, models and execution are spread across many independent participants, AI can become closer to open infrastructure than another corporate monopoly.” Levkin further lauded the decentralized approach by adding that “it also creates better conditions for privacy, encryption, auditability and local control.”
History repeats itself
The path of centralized AI development could have ramifications well beyond the technology world, with parallels to the colonial expansion of the past few centuries by the British.
The British Empire expanded its reach throughout the known world from the 16th century through well into the 19th by building a global integrated infrastructure of railways, ports and shipping lanes that other nations had to rely on. The U.S. and China appear to be conquering the digital realm in the same way through cornering the AI market by owning the compute, models, and APIs used throughout the world.
“Ultimately, the question of individual rights is being arbitrated through the current systems in place. ”
When asked about this hyper-reliance of other nations on AI-first countries, Lopez reframed the question: “Colonial is the wrong metaphor. It’s closer to the nuclear umbrella. Most nations never built nuclear weapons, but they organized their entire security posture around the two countries that did.”
Similar to the AI production imbalance, the United States and Russia dominate nuclear weapon production, while other countries are forced to adjust their policies in lockstep with global leaders. “If AGI stays locked in the US or China, the other 190 nations don’t become colonies, they become alliance shoppers,” Lopez said. He further drew a historical parallel: “They’ll pick a stack the way they picked sides in the Cold War and the switching costs will be enormous.”
If this polarizing scenario plays out, governments around the world will be faced with the political ramifications of their technology alliances. This presents myriad challenges for countries outside the US and China, chiefly the question of how governments can regulate foreign AI systems that manifest AGI agents capable of operating as business, political, or intelligence actors.
David Meldofsky, a California Attorney and editor of Lawsuit Intelligencer, told AGI Ethics News that governments are currently regulating what they can, but the groundwork is still being laid in most places. “Governments can’t inspect the model, but they can regulate the points of contact, including app stores, payment rails, API access and the domestic companies that deploy it … that’s how the US already handles foreign software it distrusts.”
For countries outside of the major AI industries, Kenneth Eade, an e-commerce attorney and manager at Amazon Sellers Attorney, emphasizes conditional market access in AGI regulation. “The EU AI Act already does this. It never inspects American model internals,” Eade said in an email to AGI Ethics News. “It conditions the legal right to operate in the EU on compliance and requires foreign providers to appoint an in-jurisdiction representative who can be served, sued and fined.”
Eade says governments have the option to block access altogether. “That’s the correspondent-banking and the TikTok model. It’s crude, it fragments the internet and it’s the fallback every regulator reaches for when the other tools fail.” The only downside of such a regulation is that it’s preventative and does nothing to ensure access for those who otherwise would be reliant on the whims of a foreign government.
Ultimately, the question of individual rights is being arbitrated through the current systems in place. In the EU, for example, users have many privacy and personal data protections, but access to models from the US and China cannot be guaranteed. And, in China, where hardware resources are being consolidated and the government oversees development with a steel gaze, such user protections are non-existent but many models, such as the popular DeepSeek, are released without global oversight.
Ethical concerns addressed in this article:
- How can cross-border regulatory principles for AGI be harmonized given competing national interests and legal systems?
- Could accelerated progress in wealthy nations exacerbate global inequality, and what interventions are needed to support the developing world in sharing AGI’s benefits?
- Who should have access to AGI technologies, and how do we prevent their monopolization by wealthy corporations or governments?
Get sharper analysis on AGI ethics
Original essays, expert commentary, and curated analysis on the human stakes of AGI.
Venus Kohli is an engineer turned freelance technical writer covering electronics and enterprise IT. Her bylines appear across TechTarget, ITPro, and AspenCore. She truly believes in the power of AI, eventually AGI. When she’s not writing content, she’s probably generating AI images of deep space, black holes, and the edge of the universe.