Tencent Chief AI Scientist Yao Shunyu says AI's bio abilities will soon make it "possible to do harm" and Anthropic has a "strong argument" for closed models

AI models can now generate actionable biological protocols, shifting from theoretical risks to practical concerns for biosecurity.

Tencent's Chief AI Scientist, Yao Shunyu, stated that AI's biological capabilities are rapidly advancing and will soon reach a level where they can be used to cause harm. This development is occurring as AI is increasingly applied to medicine and biology, promising accelerated drug discovery but also raising concerns about the potential misuse for bioweapons or initiating pandemics, particularly with open-weight models. Yao emphasized the responsibility of AI companies, including those in China, to address these security and safety issues. Tencent itself is exploring AI for scientific research, having recently released its Hy4 model, which has undergone safety evaluations. Yao also discussed the debate around open-weight versus closed-source AI models, noting that while closed models, favored by companies like Anthropic, offer tighter control, open models can contribute to defense strategies. He suggested a hybrid approach might be optimal, balancing empowerment with security.

The rapid advancement of AI in mastering biology signifies a critical inflection point, moving beyond theoretical risks to tangible capabilities that could be weaponized or misused. While this progress holds immense promise for scientific discovery, it simultaneously elevates the urgency for robust safety protocols and responsible deployment strategies, especially as the lines between defensive and offensive applications blur. The debate over open-source versus closed models intensifies, as different approaches present distinct risks and benefits in managing these powerful new capabilities. The ability of AI to understand and manipulate biological systems on a fundamental level presents a dual-use challenge that mirrors, and potentially surpasses, the complexities seen in cybersecurity. The convergence of AI with biology necessitates a proactive global approach, involving not just technical safeguards but also societal and organizational frameworks to mitigate potential harms. Future developments will likely focus on the efficacy of containment strategies for advanced models and the establishment of international norms for AI in sensitive scientific domains.