Tencent’s chief AI scientist, Yao Shunyu, warned on November 8, 2026, that artificial intelligence models are approaching a critical threshold where they will possess the technical capability to inflict biological harm.
The assertion, delivered at a science symposium in Hong Kong, highlights the mounting security risks inherent in increasingly sophisticated frontier AI systems.
The warning arrives just days after the Shenzhen-based tech giant released its latest flagship model, Hy4, on November 4. Speaking to an audience of computer science students, Yao emphasized that the transition from cybersecurity threats to biological risks changes how the industry approaches safety.
According to the South China Morning Post, Yao stated, “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.”
Tencent Curates Training Data to Prevent Malicious Use
Tencent has attempted to mitigate these risks within its own development pipeline by curating the training data for its newest model. During the launch of Hy4, the company announced that it intentionally excluded material related to software exploit development. Instead, the model was trained exclusively on defensive security data, a move intended to prevent the tool from being used for malicious cyber activities.
The company maintains that while Hy4 possesses a robust foundational knowledge of biology, its capacity for experimental design remains significantly lower than that of top-tier global models.
The Complexity of Biological Risk Management
Yao, who transitioned to Tencent last year following a tenure at OpenAI, argued that biological risks present a unique challenge that may prove more difficult to contain than cyber threats. While cyber and biological capabilities are both inherently “dual-use”—serving both attackers and defenders—Yao noted that disentangling these functions is significantly more complex in the biological domain.
The potential for these systems to accelerate drug discovery is substantial, yet the same capabilities could theoretically be repurposed to create bioweapons or trigger public health crises. Yao emphasized the necessity for rigorous evaluation processes, noting that Tencent is actively collaborating with specialized internal security labs to refine safety protocols.
As reported by the South China Morning Post, he stressed that firms releasing open-weight models carry a heightened responsibility, as these systems can be downloaded and modified by any user with sufficient hardware resources.
Open Versus Closed Model Governance
The industry remains deeply divided over the merits of open-weight versus closed-source deployment strategies. Yao acknowledged that US-based companies like Anthropic have presented a compelling case for keeping frontier models closed to prevent widespread societal harm. However, he maintained that a hybrid approach might be more effective, arguing that open models provide essential tools for researchers and defenders working to secure the ecosystem.
Tencent plans to release the model weights for Hy4 alongside a detailed report on its internal safety testing in the coming days, according to Reuters. This transparency is intended to address concerns regarding the safety of open-weight software. The following table outlines Tencent’s reported safety measures and capabilities regarding the Hy4 model:
| Feature | Status |
|---|---|
| Offensive Cyber Training Data | Excluded |
| Defensive Cyber Training Data | Included |
| Biological Experimental Design | Below high-risk threshold |
| Model Release Format | Open-weight |
Future Implications for Autonomous Labs
Looking beyond the immediate safety debates, Tencent is moving toward the implementation of autonomous science labs. These systems are designed to automate complex, end-to-end research processes, such as inverse protein folding, with minimal human intervention. Yao characterized the remaining hurdles to this automation as primarily organizational and societal rather than purely technical.
For now, the industry watches as Tencent prepares to make its weights public, testing the theory that openness can serve as a viable component of a long-term defense strategy.
