According to Anthropic concerns, three AI companies, DeepSeek, Moonshot, and Minimax, were found to copy abilities from another software, ClaudeAI. As claimed by the accuser, they sent approximately 16 million messages to Claude by creating 24,000 fake accounts.
Recently, China’s restrictions on Nvidia chips have explained the escalation between the United States and China. The tech war between Washington and Beijing has entered a new phase as the US government is planning to restrict the updated models of AI semiconductor chips to China.
Governments such as the US plan to control the export of advanced AI chips. But the continuous availability of AI software online is creating regulatory and security challenges.
Anthropic’s Claim
Anthropic is a US-based AI research and security company that founded Claude. They accused “foreign agents” of trying to enter their AI software to gather information.
Reports from Reuters and TechCrunch show growing concern about security risks. However, the AI industry remains wary that Chinese AI teams might be experimenting with or secretly analyzing the Western AI models.
What is mining?
Mining is the concept of gathering insights from AI software without formal access to build another system. Distillation and large-scale automated querying are the most common methods.
Distillation is about utilizing the information from a strong AI system and copying it to a smaller model. A large-scale automated querying model is all about sending AI multiple prompts to figure out its pattern or study its behavior.
Anthropic sees this as a “risk,” as their systems are built on updated data transfer models, a large amount of information, and advanced graphics processors, which shows a million-dollar investment and a decade of research.
US Approach: Balancing Innovation and National Security
Recently, the US government implemented export restrictions on advanced AI chips, such as those from AMD and NVIDIA, to China. These chips increase the efficiency of AI models while training at much cheaper costs.
In the absence of an advanced semiconductor, the cost of production rises exponentially, with an adverse effect on the speed.
According to the Anthropic blog post, the illegal distillation of US-based AI models by foreign companies will be a grave security threat.
Replicating the models without proper protection will cause misuse, such as a cyberattack or the development of bioweapons. However, such issues are mitigated in Anthropi’s case due to robust security measures, Antropic said.
China’s AI Landscape
China has seen vast AI developments in the last decade. The semiconductor restrictions from the US have not stopped China from building updated AI models. Tech giants like Alibaba, Baidu, and Tencent roll out their AI models.
Western AI models can accelerate the speed, and concerns exist, but there is no public evidence that confirms that Chinese firms have illegally replicated Claude.
The US is leading the AI market with high-end chips and top-tier AI models. Meanwhile, China is pursuing powerful AI systems despite limited access to advanced hardware.
Future Outlook
The US-China AI competition is rising, along with concerns for national security and market competition. It is still uncertain whether the United States will respond with broader control and restrictive AI model policies.
The updated rules may slow down the innovation, as research centers, universities, and businesses may find it difficult to share information across borders.
The next phase will be about the “security” of the AI system. As for governments, it is far easier to oversee physical AI hardware, but now they have to focus on how it can manage access to online AI models.
Source: Reuters, TechCrunch, Anthropic Blog