Thought Leadership
A bipartisan initiative in the US Congress is seeking to ban artificial intelligence (AI) from China, Russia and other countries considered hostile for use in US authorities.
The bill, known as the “No Adversarial AI Act”, was introduced in both the House of Representatives and the Senate.
Security concerns: AI from authoritarian states in our sights
The aim of the law is to create a clear distinction from AI systems that are under the influence of authoritarian governments. Criticism focuses in particular on Chinese models such as DeepSeek, but also on developments from Russia, Iran and North Korea.
According to the initiators of the law, these technologies pose a significant risk to national security. The concern is that the systems could intercept data, support surveillance or be misused for cyber attacks.
The background: AI as a key strategic technology
John Moolenaar, Republican MP from Michigan and Chairman of the Special Committee on the Influence of the Chinese Communist Party, warned against the military use of AI by China. He is supported in this by Democratic MP Raja Krishnamoorthi. Both see AI as a crucial technology in geopolitical competition – especially in the context of an intensifying technological arms race.
DeepSeek in particular is a prime example of this threat. The company published its first major language model in November 2023. It now claims to have surpassed Western systems such as ChatGPT or Meta LLaMA in terms of performance and cost with its new DeepSeek-R1 model. However, critics point to its close ties to the Chinese government and a data protection policy that provides for the storage of user data in China.
Concrete measures of the law
The draft law includes several measures:
- Prohibition of the use of certain AI models by federal authorities if they originate from so-called “adversary states”.
- Obligation to publish an official list of prohibited AI models by the US Federal Acquisition Security Council – updated every 180 days.
- Exemptions only within narrow limits, for example for research purposes or national security concerns. It must also be proven that the respective model is not under the control of a foreign adversary.
Deeper problems with Chinese AI developments
In addition to DeepSeek, Manus – an AI assistant released in March 2025 by the Beijing start-up Butterfly Effect – is also attracting attention. Manus is considered more advanced than many Western chatbots, including ChatGPT, Claude and Grok. At the same time, Baidu launched its own model, ERNIE X1, on the market.
US researchers also warn that Chinese models such as DeepSeek-R1 could be misused to develop malware – another argument in favor of a legal ban in the government sector.
Reactions and further steps
Co-signers of the bill include Congressmen Ritchie Torres (D-NY), Darin LaHood (R-IL) and Senators Rick Scott (R-FL) and Gary Peters (D-MI). According to a Reuters investigation, DeepSeek is said to have already had access to large quantities of Nvidia chips and to have helped build up Chinese military capacities.
The safety of AI models from abroad is the subject of intense debate in the USA – especially since the first US authorities have banned the use of such systems for fear of data misuse.
With the “No Adversarial AI Act”, US lawmakers want to send a clear signal: Technologies that are potentially in the service of authoritarian regimes should have no place in the infrastructure of American federal authorities. The debate shows the extent to which AI has become a key geopolitical technology – and the extent to which security issues determine its use.
