Tencent scientist warns AI biology gains could enable bioweapons

By Tom Whitfield

Tencent released its newest large model, Hy4, on Nov. 4, saying internal tests found it could not meaningfully make harmful biological experiments easier. That company finding came as Tencent's chief AI scientist warned that advanced AI could soon master biology and turn cyber or biological capabilities into a global security risk.

What is Yao Shunyu warning about?

Yao Shunyu, a former OpenAI researcher, said AI models will soon be able to master biology. He named biological weapons and new pandemics among the risks, according to the South China Morning Post, which reported his remarks at a science symposium organised with the Hong Kong Academy of Sciences and the Future Science Prize.

The symposium marked a rare public appearance by Yao. He compared the coming shift to the alarm over cybersecurity that followed AI's leap in coding ability. He said the models' cyber and biological capabilities are approaching a threshold of tangible harm, while open-weight models used without regulation could multiply the risks.

Yao called on Chinese and international developers of open-source and commercial models to act responsibly. His warning describes a future capability threshold, while Tencent's Hy4 result describes a present internal test. The practical takeaway is simple: controls need to arrive before advanced models can cause tangible harm.

What did Hy4's safety testing show?

Tencent limited Hy4's cybersecurity training to defensive work and left out material on building hacking tools and breaking into systems, Reuters reported. The choice placed a deliberate boundary around the model's cyber capabilities.

The company said its testing found Hy4 incapable of meaningfully making harmful biological experiments easier. Tencent also plans to release the model's weights and a report on its safety testing. That would put both the model and the stated basis for its safety conclusion within reach of outside scrutiny.

“We recognise the potential safety risks posed by advanced cyber and biological capabilities,” Tencent said. Until the weights and report appear, the safety conclusion remains a company-reported negative test result rather than a completed public evaluation. The practical takeaway is that the negative finding has not yet faced outside scrutiny.

What does Tencent want from autonomous laboratories?

Tencent has an ambition to build autonomous science laboratories that can run complex biological processes end to end, including reverse protein folding. Yao said the technical problems could be solved over time, but the harder challenge would be how society and organisations adapt. The lab effort is a stated ambition, not confirmed operating capacity.

His framing separates engineering progress from institutional adaptation. He expects the technical obstacles to become solvable over time, while society and organisations must adjust to the capabilities that emerge.

The result is a roadmap in which laboratory automation and social adaptation must move together. In plain terms, Tencent's long-term biology goal depends as much on institutional readiness as on solving the technical steps.

Tencent named Yao chief AI scientist in December 2025, and he reports to president Martin Lau. His role links the broader risk warning to the company developing Hy4 and pursuing laboratory automation.

Tencent has already limited Hy4's cyber training, while its safety report remains pending and autonomous laboratories remain a future objective. In practical terms, only the defensive cyber restriction is already in force.

How could open weights change the risk?

Yao said open-weight models used without regulation could multiply the risks. His call for responsibility covered Chinese and international developers of both open-source and commercial models, so the issue reaches beyond Tencent's own release. The immediate commercial question is how freely other developers can adapt Hy4 once Tencent releases the planned weights.

The planned Hy4 weight release and safety report will be the next measurable checkpoint. They would move some scrutiny outside Tencent and let developers compare the model with the company's account of its safeguards. Yao's concern rests on access and regulation as much as on model capability.

Developers will decide whether stronger safeguards, or the regulatory gaps Yao describes, shape competition between open-source and commercial models. The forward-looking commercial test is whether Hy4's planned weight release makes its safety controls a standard, or whether unregulated use produces the proliferation Yao warns about.