Anthropic Alleges Chinese AI Firms Compromise US National Security
Synopsis
Anthropic has raised alarms over three Chinese AI companies allegedly stealing capabilities from its Claude model. The firm warns this could pose serious national security risks, enabling authoritarian regimes to misuse advanced AI technology.
Key Takeaways
Anthropic claims illegal extraction of AI capabilities by Chinese firms.
Concerns raised over national security and misuse of AI.
24,000 fake accounts reportedly used for training models.
Models may lack safety features, posing risks for cyberattacks.
US export controls questioned in light of these events.
New Delhi, March 25 (NationPress) The US-based AI company Anthropic has accused three Chinese tech unicorns—DeepSeek, Minimax, and Moonshot AI—of unlawfully acquiring capabilities from its Claude model, as detailed in a recent report.
The CNN Business report highlighted that the firm raised concerns about national security due to the alleged theft, which involved a method known as distillation.
According to the allegations, the Chinese entities created approximately 24,000 fake accounts to train their models, utilizing over 16 million interactions with Claude.
Anthropic cautioned that AI models developed through these means might lack the necessary safety measures implemented by responsible companies, potentially making them tools for cyberattacks and biological warfare.
These models could empower “authoritarian regimes to utilize advanced AI for offensive cyber tactics, misinformation efforts, and extensive surveillance,” the company stated, emphasizing that “the opportunity to intervene is limited.”
CNN has reached out to DeepSeek, MiniMax, and Moonshot AI for their responses, according to the report.
DeepSeek’s rapid rise in China, gaining the label of “AI tigers,” has raised concerns that US export regulations have fallen short.
The trio of unicorns currently ranks among the top 15 models on the renowned Artificial Analysis leaderboard, as noted in the report.
Nevertheless, Anthropic contends that these distillation attempts demonstrate the effectiveness of export controls, asserting that advanced model development cannot continue without access to high-tech chips.
Similar accusations were previously made by OpenAI, which claimed that DeepSeek has been “free riding” on the advancements created by OpenAI and other leading US laboratories.
Recently, Anthropic PBC was officially designated as a “Supply Chain Risk (SCR)” by the US government, and the CEO of the firm apologized for prior criticisms directed at President Donald Trump.
The company clarified that this designation pertains solely to the use of Anthropic’s Claude models in Department of War contracts, not extending to “all uses of Claude by clients holding such contracts.”
Frequently Asked Questions
What allegations has Anthropic made?
Anthropic has accused three Chinese companies of illegally extracting capabilities from its Claude model, raising national security concerns.
What method was reportedly used for the alleged theft?
The alleged theft was conducted through a process called distillation, utilizing fraudulent accounts to train models.
What risks do the models produced by these firms pose?
The models may lack safety measures, potentially being used for cyberattacks and biological warfare.
What is the significance of this situation?
It highlights the ongoing challenges of protecting technological advancements and the implications for national security.
What has been the response from the accused companies?
As of now, CNN has reached out for comments from DeepSeek, MiniMax, and Moonshot AI, but responses have yet to be reported.
