Open AI models could enable bioweapon development, Tencent scientist warns

Open-weight artificial intelligence models could soon become capable of causing serious harm in biology and cybersecurity, Tencent chief AI scientist and former OpenAI researcher Vinces Yao Shunyu has warned.

Speaking at a science symposium in Hong Kong on Saturday, Yao said AI models were approaching a threshold where their capabilities could be misused.

“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,” he said.

While AI can help smaller laboratories and independent researchers with scientific research and drug discovery, open-weight models could also make dangerous applications, including biological weapons development, more accessible to malicious actors.

Yao said biological capabilities could be harder to regulate than cybersecurity because defensive and offensive applications are difficult to distinguish. He urged AI companies to take security and safety concerns seriously.

Tencent is working towards autonomous science laboratories where AI systems could independently conduct complex experiments, including protein-folding research.

Its latest flagship model, Hy4, is designed to balance accessibility and safety. The company said it had removed the model's hacking capabilities while retaining cybersecurity defence functions. Its biological capabilities remain below the company's high-risk threshold.

Tencent plans to release Hy4's model weights along with a system card detailing its safety evaluations.

Yao said neither completely open nor entirely closed AI models might be ideal, arguing that a combination of both approaches could allow wider access to technology while addressing security risks.