Tencent's chief AI scientist has warned that artificial intelligence models are closing in on the ability to master biology, bringing with them risks that could prove harder to contain than the cyberattacks the industry already grapples with.
Vinces Yao Shunyu, Tencent's chief AI scientist and a former OpenAI researcher, issued the warning on Saturday at a science symposium in Hong Kong co-hosted by the Hong Kong Academy of Sciences and the Future Science Prize. It was a rare public appearance for the 29-year-old, who joined Tencent last year after a one-year stint at OpenAI, where he contributed to the company's Operator and Deep Research products.
"The cyber abilities of [AI] models and soon [the] bio ability of models will reach a threshold where it's actually possible to do harm," Yao told the gathering. "All the Chinese or any open model company needs to be very responsible and very serious about those security and safety concerns."
A threshold where real harm becomes possible
Yao's central argument is that biology is the next major frontier for AI, following coding — a field where AI has already transformed cybersecurity, for better and worse. As frontier systems are increasingly applied to medicine and biology, they could accelerate the discovery of new drugs and treatments. But the same capabilities could also make it easier for malicious actors to design biological weapons or, in the worst case, help trigger a new pandemic — especially if the models are released as open-weight software that anyone can download and run.
He warned that AI's biological capabilities may be more "complicated" to manage than its cyber capabilities, in part because defensive and offensive applications are harder to separate in biology than in software. A model that can help a legitimate researcher design a new therapy could, with little modification, help someone else design something harmful.
Hy4 and the open-weights test
The warning carries extra weight because it comes from the person responsible for the model in question. Tencent released Hy4, its latest flagship AI model, on Wednesday. Benchmarking firm Artificial Analysis ranked it the second-best performing model produced in China, and Tencent is exploring using AI to automate scientific research.
Tencent says it took deliberate steps to limit Hy4's misuse potential. The company removed the model's offensive cyber capabilities while retaining its defensive ones by filtering relevant data during training. It also says Hy4's biological capabilities remain "well below" its high-risk threshold, because the model still cannot turn foundational biological knowledge into actionable protocols.
The company has pledged to release Hy4's weights in the coming days, along with a comprehensive system card setting out its cybersecurity and biological safety evaluations. That disclosure will let outside researchers audit the model and test whether Tencent's safety claims hold up — an early test of how seriously an open-weight lab takes the risks its own chief scientist is flagging.
An industry-wide question, not just Tencent's
Yao's warning fits a wider pattern. Tencent has already signed on to the view that frontier AI safety is a public-health question: together with Huawei and Alibaba, it contributed to a frontier safety report at the recent Wuzhen summit that named pathogen design as a capability that could create major public health risks.
For now, the gap between what models can do in biology and what would be needed to cause real harm still exists — but Yao's message is that it is closing. The question he left with the industry is whether the safety infrastructure around open models is being built as fast as the capabilities themselves.
By Hamza Faisal
