AI models will soon be capable of causing harm in biology, similar to how they already can in cybersecurity, according to Yao Shunyu, the chief AI scientist at Tencent.[1][2] Yao spoke at a science symposium in Hong Kong, the South China Morning Post reported.[2][1]
Yao said Chinese companies that release open models, which anyone can download and change, need to take safety very seriously.[1] He also warned that biology risks may be harder to manage than cyber risks, and AI Chat Daily reported he said attack and defense are easier to separate in cybersecurity than in biology.[1][2]
The warning follows the Nov. 4 launch of Hy4, Tencent's new flagship model.[1][2] Artificial Analysis, a benchmarking firm, ranks Hy4 as the second-best model produced in China, the South China Morning Post reported.[1][2] Tencent said it trained Hy4 on defensive security data and left material on exploiting software flaws out of its training data.[1] Tencent said internal expert reviews show Hy4 has strong basic knowledge but is weaker than leading models at designing experiments; the company stated the model's biology capability sits well below its high-risk threshold.[1][2]
Yao said US firms like Anthropic make a strong case for closed models, but argued a mix of open and closed models may work best because open models can help defenders.[1][2] He pointed to Hugging Face reaching for a Chinese open model for cybersecurity after American closed models declined the task on guardrail grounds, Tech360 reported.[3]
Tencent plans to release Hy4's weights and a report on its safety testing in the coming days, Reuters reported.[1] Yao said the evaluation process with Tencent's security labs is not yet complete.[2]