DeepSeek engineer defends push for open-source AI smarter than humans despite higher risks

DeepSeek engineer Gao Xi defended the company's push for open-source AI that could surpass human intelligence, despite increased safety risks, during a recent AI panel at the World Internet Conference in Wuzhen, China. Gao argued that open-source development fosters shared innovations and technical breakthroughs, and that the rapid progress of AI should be accelerated rather than slowed down by rivals.

He emphasized that time is the main competitor in the AI race, and that technical barriers from competitors would only temporarily slow progress, not stop it entirely. Gao's comments come amid growing safety concerns in the US and scrutiny in China over potential national security risks posed by open-source AI. Open-weight models, which allow unrestricted access and modification of AI code, have been championed by Beijing as a means to counter US tech containment.

However, security experts warn that releasing model weights could lower the barrier for malicious actors to launch cyberattacks or develop bioweapons. Gao acknowledged that open-source releases may exacerbate safety issues, but stated that DeepSeek is working to strengthen safety features, such as model classifications, though he did not disclose specific details.

Alibaba Group's project lead Liu Dayiheng acknowledged the additional safety challenges posed by irreversible open-weight models, but advocated for a comprehensive governance regime and industry standards as a solution, rather than technological gatekeeping. Alibaba Group owns the South China Morning Post. China is balancing the safety concerns with the push for open-source innovation, proposing new export controls targeting frontier AI models with advanced cyber-offensive capabilities while supporting open-source software.