Anthrop may be in the midst of an existential legal battle with the US government, but that hasn’t stopped it from continuing to consider how the US should deal with China when it comes to AI.
In a 5,500-word policy paper posted Thursday, Anthropic issued a dire warning to U.S. leaders: The United States must do everything in its power to prevent China from overtaking China in AI.
democratic AI
Anthropic depicts two possible scenarios.
1. 🇺🇸 🥇 US maintains 12-24 month lead over Chinese AI
In the first scenario, where the United States maintains its lead, a combination of strict export controls on advanced AI chips, protection from “distillation attacks,” and industry-wide transparency will help build a bulwark against China’s competition to build strong domestic AI.
The paper argues that it would give the United States a military and national security advantage over its greatest adversaries, and would have the effect of attracting top AI talent from around the world.
Industry leaders can set standards for transparency and safety, and it’s critical that the U.S. stands in that position, the newspaper said.
2. 🇨🇳 ⚖️ China catches up with US-made frontier models
If American companies consider China just another market for their advanced AI chips, Anthropic sees a number of negative consequences for the world. China could now close what experts see as a 12- to 24-month gap between its top-of-the-line models and American-made Frontier models.
China’s “AI-enabled technological authoritarianism” could be used to increase mass surveillance, reinforcing already tight controls on the population.
Anthropic, which just warned the world about the potential cyber risks posed by its upcoming Mythos model, is concerned that such advanced tools could be used for offensive cyberattacks.
AI is accelerating research and development, and more capable models will help China make faster progress in all fields. China is focused on integrating domestic AI technology into society to accelerate progress, and even if it does not have the most advanced models, it can export its tools around the world.
“If the Chinese Communist Party more quickly and effectively integrates nearby AI systems into the Chinese economy and its security apparatus, and promotes the global adoption of subsidized, low-cost AI, it could secure an advantage over democracies to overcome intelligence deficiencies.”
Export controls work
The term “export control” appears 16 times in Anthropic’s paper and spends much ink discussing its effectiveness and necessity.
Cutting-edge chips (e.g. Nvidia —China continues to be at a disadvantage as it struggles to compete with advanced chip manufacturing and design from liberal democracy-based companies. ASML(Netherlands), AMD (us), TSMC(Taiwan), Samsung (South Korea). Anthropic argues that the innovations these companies create can only occur in democracies.
China still has a ways to go when it comes to chips. There is reportedly a working prototype of a chip factory employing extreme ultraviolet lithography (EUV), a technology currently mastered by only one company. Faster chips bring more powerful AI.
What grows in the shadows
Anthropic, OpenAI, and other AI labs have a tradition of releasing detailed and transparent technical documentation that explains their capabilities and detailed safety risks. The paper points out that more and more Chinese model releases are not reaching this level of transparency and are starting to keep the models exclusive. This could significantly increase the risk that AI models could be used to develop chemical, biological, or nuclear weapons.
President Trump has just returned from a visit to China with Nvidia CEO Jensen Huang. It is not yet clear what was discussed in relation to AI and chips. But for now, Anthropic will keep its eyes on the rearview mirror as the Chinese model improves.
