Tencent's chief AI scientist, Yao Shunyu, has warned that AI models will soon be capable of causing harm in biology the way they already can in cybersecurity, and that the industry is not ready for it. Yao, 29, spoke on Saturday at a science symposium in Hong Kong jointly hosted by the Hong Kong Academy of Sciences and the Future Science Prize, the South China Morning Post reported. It was a rare public appearance for the former OpenAI researcher, who joined Tencent last year.
The remarks came days after Tencent launched Hy4, its new flagship model, on November 4. Benchmarking firm Artificial Analysis ranks Hy4 as the second-best model produced in China. Tencent has said it will publish Hy4's weights alongside a system card covering cyber and biological evaluations in the coming days.
Yao's argument runs in sequence: cyber capabilities first, biology next. He said model abilities in both areas will reach a threshold where real harm becomes possible, and that Chinese companies releasing open models need to take safety very seriously. He also said the gap between Chinese and American models is not large.
Key facts
- 01Tencent chief AI scientist Yao Shunyu spoke Saturday at a Hong Kong science symposium about AI biology risks.
- 02Yao, 29, joined Tencent last year after a year at OpenAI.
- 03Tencent launched its flagship Hy4 model on November 4; Artificial Analysis ranks it the second-best model made in China.
- 04Tencent says Hy4's biology capability sits well below its high-risk threshold.
- 05Tencent plans to release Hy4's weights with a system card covering cyber and biological evaluations.
Yao singled out biology as the harder problem. In cybersecurity, he said, attack and defense are easier to separate than in biology, which is why he wants careful evaluation of biological capabilities before a model is released.
Tencent's handling of cyber risk in Hy4 shows the approach it has taken so far. The company left material on exploiting software flaws out of Hy4's training data and trained it on defensive security data instead, aiming to strip offensive hacking ability while keeping the defensive side.
On biology, Tencent says Hy4's capability sits well below its high-risk threshold because the model still cannot turn textbook biology into an actionable protocol. Based on internal expert reviews, Tencent says Hy4 has strong basic biology knowledge but is weaker than leading models at designing experiments. Those are Tencent's own assessments; the forthcoming system card is where the company has promised the detailed evaluations.
Yao said Tencent is working with its security labs on evaluation, but that the process is not yet complete.
He did not come down on one side of the open-versus-closed debate. Yao said US companies such as Anthropic make a strong case for keeping top models closed, but argued that a mix of open and closed models may work best, because open models can also help defenders. He cited Hugging Face using a Chinese open model to defend against a cyberattack after closed American models refused to help.
Tencent, along with Huawei and Alibaba, contributed to a frontier AI safety report presented at the Wuzhen summit that named pathogen design as a capability with the potential to create major public health risks.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
