In the rapidly intensifying global competition for artificial intelligence (AI) supremacy, Michelle Askin, CEO of Reflection, a company developing open-source AI models, has voiced concerns about the current trajectory. In a recent interview, Askin underscored China's dominance in the realm of open AI models and emphasized the critical need for the United States to establish a competitive foothold in this sector.

China’s lead in open AI models

According to Askin, the best open-source AI models available today are coming out of China, and these models are being increasingly adopted by nations and enterprises worldwide. This adoption trend is not just about AI models themselves but also about the infrastructure and ecosystems tied to them. “Entire nations and enterprises today are thinking about how to not just rent intelligence but own it,” said Askin. The reliance on open-source models enables them to build their AI capabilities in-house, ensuring independence and control.

This adoption creates a significant strategic advantage for China. When governments and organizations turn to Chinese open models, they often become reliant on the broader infrastructure provided by Chinese companies, which can include hardware, chips, and software. In Askin’s words, “It’s not just about the models; it’s the entire stack and ecosystem.” This, she suggests, is the real race between the US and China — not simply the sophistication of the models, but the long-term dependencies created by the surrounding technology stack.

Reflection’s ambition to provide a counterbalance

Reflection is positioning itself as a counterweight to China’s dominance in open AI. Backed by a recently closed funding round reportedly worth $2 billion — which valued the company at $25 billion — Reflection plans to build advanced American open models that could supply sovereign AI deployments. Among its investors are notable institutions like JP Morgan, which joined through its Strategic Initiative Group led by Todd Combs, formerly affiliated with Berkshire Hathaway.

The company, comprised of over 150 AI experts, is still working on its first flagship model, which has yet to be publicly released. However, Askin expressed confidence in the model’s progress and the company’s ability to compete. “These things are hard to build. They’re kind of like rocket ships,” she noted, highlighting the complexity and precision required to develop competitive open AI systems. While acknowledging that Reflection’s work is still in progress, the company’s ability to secure sovereign deals even in this pre-launch phase suggests strong confidence in its potential.

Open-source AI: a Trojan horse?

One of the most compelling arguments put forth by Askin centered on the strategic implications of open-source AI. She described open-source models as “Trojan horses” for the broader infrastructure they bring. Countries and enterprises opting for open models often become reliant on the surrounding ecosystem — an area where Chinese technology providers currently have the edge. By offering competitive open models, Reflection aims to attract nations and companies that value independence but also wish to avoid locking themselves into China’s tech stack.

While closed-model providers like OpenAI, Anthropic, and Alphabet pursue proprietary systems, Reflection is banking on the open-source approach to appeal to those who prioritize autonomy and ownership. Askin noted that within the sovereign deals Reflection is pursuing, the competition is less with closed-model providers and more against Chinese open models.

Safety and risks of open AI

The broader industry debate over the safety of open-source AI was another critical topic in the interview. Askin acknowledged the concerns that greater access to powerful models could allow malicious actors to misuse them. Yet, she argued that open models can also enhance safety by fostering an ecosystem of developers and researchers who actively participate in identifying and mitigating risks.

“There are two perspectives,” said Askin. “One is that these models should only be available to a select few safety researchers. The other is the belief that the best and safest models come from having a broader community involved in their development and oversight.” Reflection aligns with the latter viewpoint, advocating for open collaboration to ensure that safeguards evolve alongside technological capabilities.

This stance contrasts with companies like Anthropic, which chose not to release its Mythos model due to safety concerns. The decision highlights a fundamental divide in how industry leaders envision the proper balance between innovation, accessibility, and security in AI development.

The US-China AI rivalry and the road ahead

The implications of Askin’s comments stretch well beyond Reflection itself. The ongoing AI race between the US and China is not merely a competition for technological breakthroughs; it is also about shaping the global AI infrastructure. China’s ability to lead in open models poses a challenge to US dominance, not just in AI innovation but in technological influence more broadly. In this context, Reflection’s mission to build competitive open-source models on US soil takes on heightened geopolitical significance.

Still, developing these systems is no small undertaking. Askin’s rocket ship analogy reflects both the ambition and the technical hurdles involved. To succeed, Reflection must deliver not just a powerful open model but also a robust ecosystem that rivals what Chinese providers offer.

Conclusion

Reflection’s pursuit of open-source AI leadership highlights critical dynamics of the global AI race, particularly the strategic tension between openness and control. By prioritizing open models, Reflection aims to empower nations and enterprises to take ownership of their AI capabilities. However, as the competition heats up, the broader industry will continue grappling with issues of safety, accessibility, and the long-term implications of open-source development.

In the high-stakes race between the US and China, the outcome may hinge on which country can better align technological innovation with global adoption. For Reflection, and the broader US AI community, the race is not just about winning but about redefining what winning means in the age of AI.