Tencent AI Scientist Warns of Rising Biological Risks from Advanced Models

Floating metallic cubes in shades of blue and purple with reflective surfaces — Milad Fakurian / Unsplash

Yao Shunyu, a chief AI scientist at Tencent and a former researcher at OpenAI, stated that frontier artificial intelligence models are rapidly advancing into the field of biology. Speaking at a scientific symposium in Hong Kong, he indicated that these systems are approaching a critical threshold where their capabilities could cause real-world harm. This development has triggered significant concern within the industry regarding the emergence of new biological safety challenges.

The scientist noted that while AI has already demonstrated capabilities in cybersecurity that can produce actual impact, its biological capabilities are now approaching a similar level of proficiency. He warned that managing the development of AI in biology may prove more difficult than managing its growth in cybersecurity. Consequently, he urged both Chinese enterprises and other companies following an open model route to treat these security issues with the utmost seriousness.

In recent years, an increasing number of frontier AI systems have been applied to life sciences, medical research, and drug development. Industry consensus holds that AI is expected to significantly shorten drug screening and research cycles, thereby accelerating the discovery of treatment options. However, experts point out that these capabilities possess a dual-use nature. While they drive scientific progress, they also carry the risk of being abused by malicious actors, which could introduce new biological security hazards.

This statement coincides with Tencent's acceleration of its AI strategy. The company officially released its new flagship model, Hy4, on November 4, positioning it as one of its most important foundational models. According to data from the AI evaluation firm Artificial Analysis, Hy4 is currently ranked among the strongest performing AI models in China.

Regarding future directions, Yao revealed that Tencent is exploring the possibility of building autonomous science laboratories. The goal is to use AI systems to independently complete complex scientific research processes, including certain biological research tasks. He argued that while most technical difficulties can eventually be solved, the greater challenges lie in social governance, organizational coordination, and the construction of safety regulatory systems.

To mitigate potential risks, Tencent implemented a series of safety measures during the training of Hy4. The company stated that it deliberately excluded data related to software vulnerability exploitation and attack behaviors, retaining only defensive cybersecurity knowledge. This approach aims to preserve protective capabilities while reducing offensive potential. For the biological domain, internal assessments concluded that although Hy4 possesses strong foundational biological knowledge, there remains a clear gap before it can independently formulate experimental processes or form high-risk capabilities.

Yao emphasized that distinguishing between offensive and defensive capabilities is often more difficult in biology than in cybersecurity. This complexity means researchers face more intricate situations when evaluating model risks, necessitating stricter testing and validation before model release. Tencent is currently collaborating with its internal top-tier security laboratory to conduct systematic assessments of these capabilities.

Addressing the ongoing debate over open versus closed source models, Yao acknowledged that the view held by US companies like Anthropic, which advocate keeping the most advanced models closed to reduce social risks, has some validity. However, he also argued that open models can help defenders improve their capabilities. He suggested that the industry may need to seek a balance between openness and closure in the future, rather than relying entirely on a single approach.

As AI expands from programming, mathematics, and content generation into life sciences, biology is being viewed as one of the most important frontier directions for the next stage. Analysts point out that once AI truly possesses the ability to deeply understand and apply biological knowledge, the technology could not only transform drug research and the medical industry but also prompt global regulators to re-examine AI governance frameworks to address the resulting new security challenges.