Tencent: AI Will Soon Be Able to Help Build Bioweapons

Vinces Yao Shunyu, chief AI scientist at Tencent and former OpenAI researcher, said that soon AI could aid in developing biological weapons. Rather than keeping AI models completely locked away, he suggested removing harmful information from the training data. Tencent plans to release a new open model, Hy4, as well as a safety report.

This debate centers around controlling the risks of AI. While Anthropic keeps their models closed and filters what questions they’ll answer, Tencent wants to keep their weights open without including harmful knowledge in the training process. But filtering has its downsides — it may also prevent users from asking valid questions. Meanwhile, open models run the risk of having their safety measures removed. There is currently no consensus on a safety standard for deploying new AI models.