Tencent AI Chief Warns of Emerging Biological Risks from AI Models
Yao Shunyu highlights the potential for AI to facilitate biological harm alongside cybersecurity threats.
1Emerging Biological Threat Vectors
Tencent chief AI scientist Yao Shunyu warned that artificial intelligence models are approaching a threshold where they could facilitate biological harm. Speaking at a Hong Kong science event, Yao noted that while cyber risks are already established, the potential for AI to assist in biological threats presents a more complex management challenge.
2Open Source Safety Concerns
Yao emphasized that Chinese companies releasing open-source models must prioritize safety protocols. Because open models allow users to download and modify code, they carry inherent risks that are difficult to control. He argued that while open models aid defenders, the potential for misuse necessitates a rigorous approach to safety and oversight.
3Strategic Model Development
Tencent recently launched its flagship model, Hy4, on November 4. To mitigate risks, the company explicitly excluded material related to exploiting software flaws from the training data. Instead, the model was trained using defensive security datasets, reflecting a deliberate strategy to prevent the AI from assisting in malicious cyber activities or exploits.
4Biological Capability Assessment
Internal expert reviews conducted by Tencent indicate that Hy4 possesses strong foundational knowledge in biology. However, the company noted that the model remains weaker than leading global competitors in designing complex experiments. Tencent has not yet released the model for independent testing, though it plans to publish safety reports soon.
5Benchmarking and Model Weights
According to benchmarking firm Artificial Analysis, Hy4 currently ranks as the second-best Chinese AI model. Tencent intends to release the model's weights and comprehensive safety testing documentation in the coming days. This transparency is intended to provide insight into how the company manages the balance between performance and potential safety risks.
6Debate on Model Access
Yao addressed the ongoing industry debate regarding closed versus open models. While acknowledging that US firms like Anthropic advocate for keeping top-tier models closed to prevent misuse, Yao suggested a hybrid approach. He believes a mix of open and closed models is optimal, as open access can empower researchers to build defenses.
7Professional Background of Leadership
Yao Shunyu, 29, joined Tencent last year following a tenure at OpenAI. His background in both Western and Chinese AI development informs his perspective on the dual-use nature of generative technology. His recent comments reflect a growing consensus among industry leaders that safety must evolve alongside the rapid advancement of model capabilities.
8Future Regulatory Implications
The warning highlights a shift in focus from purely digital security to physical and biological safety in AI development. As models become more capable of synthesizing scientific information, the industry faces pressure to implement standardized safety benchmarks. Future developments will likely depend on how effectively firms can balance innovation with risk mitigation.
Original article: The Next Web — Tencent AI Chief Warns of Emerging Biological Risks from AI Models