US artificial intelligence developers accuse Chinese firms of stealing their data
US artificial intelligence developers accuse Chinese firms of stealing their data
Publish Date: 2026-02-26 05:11:00
Source Domain: www.computerweekly.com
Here is a summarized list of between 4 and 8 key points from the article:
* US AI developers, including Anthropic and Google, are concerned that Chinese labs are using distillation attacks to extract data from their models to accelerate AI development at reduced cost.
* Distillation techniques involve training smaller models based on the outputs of larger, more advanced models, which is commonly used for legitimate purposes but also has exploitative risks.
* Anthropic specifically accused three Chinese firms of creating fraudulent accounts to extract data from its AI models, outlining sophisticated techniques used for this purpose.
* Open AI also detailed how Chinese companies like DeepSeek use unauthorized resellers and advanced multi-stage processes to distill models illicitly.
* In contrast, US AI firms have been accused of using copyrighted or other data without permission to train their own models, with ongoing legal disputes over “fair use.”
* Anthropic and others defend this practice as transformative use that helps develop generative AI systems.
* There is tension between legitimate use and exploitation regarding model training data, with contrasting views on what constitutes legitimate practice between US firms and Chinese competitors.
