Tencent AI Chief Warns Models Will Soon Pose Biology Risks Harder to Manage Than Cyber Threats

Tencent AI Scientist Issues Stark Warning on Biology Risks

Yao Shunyu, Tencent's 29-year-old chief AI scientist, issued a stark warning at a science symposium in Hong Kong on Saturday that AI models will soon reach a threshold where they can cause actual harm in biology1

1

2

2

. Speaking at an event co-hosted by the Hong Kong Academy of Sciences and the Future Science Prize, Yao emphasized that these biology risks may prove harder to manage than existing cybersecurity threats. The timing of his remarks is significant, coming just days after Tencent launched Hy4, its flagship AI model, on November 4.

AI Models Pose Biology Risks Through Dual-Use Capabilities

Yao explained that frontier AI systems present a dual-use nature, capable of both advancing medical research and enabling malicious applications2

2

. While these models could aid drug discovery and treatment development, malicious actors could potentially exploit them to develop biological weapons or trigger a pandemic. "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 stated1

1

. This assessment underscores growing concerns about AI safety as models become more sophisticated in handling biological information.

Tencent Hy4 Model Trained With Defensive Security Focus

In its launch announcement, Tencent revealed that it deliberately excluded material on exploiting software vulnerabilities from Hy4's training data, opting instead to train the model on defensive security data1

1

2

2

. According to internal expert reviews, Hy4 demonstrates strong basic knowledge of biology but shows weaker experiment-design capabilities compared to leading models. Benchmarking firm Artificial Analysis ranks Hy4 as the second-best Chinese AI model. Tencent plans to release Hy4's model weights and a comprehensive safety-testing report in the coming days, demonstrating a commitment to transparency around AI safety.

Open-Weight Models Require Serious Safety Considerations

Yao specifically called out Chinese companies releasing open-weight models, which users can download and modify, stressing they must treat AI safety as a serious issue2

2

. His perspective reflects his unique position bridging Eastern and Western AI development, having joined Tencent in 2025 after spending a year at OpenAI. While acknowledging that U.S. companies including Anthropic have made compelling arguments for keeping leading models closed, Yao suggested a balanced approach1

1

. He argued that a combination of open models and closed models may work best, noting that open models can help defenders strengthen cybersecurity and biology safeguards.

Managing Cybersecurity and Biology Risks Presents Different Challenges

The distinction Yao drew between cybersecurity threats and biology risks highlights an emerging concern in AI development. While cyber abilities of AI models already pose documented risks, biology-related threats could prove more difficult to control due to their potentially irreversible and widespread consequences2

2

. This assessment suggests that existing frameworks for managing AI risks may need significant adaptation as models gain more sophisticated biological capabilities. Industry observers should watch how major AI developers respond to these warnings, particularly regarding the release strategies for open models with advanced biology knowledge and whether new safety protocols emerge specifically targeting biological applications of AI systems.