Tencent AI chief Yao: bio models near 'can do harm' threshold

Tencent's chief AI scientist used a Hong Kong science symposium on Saturday to say frontier AI is closing in on a capability line where it can cause real-world damage. "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 Shunyu said, according to the South China Morning Post.

Yao, 29, joined Tencent last year after a year at OpenAI. He said biology risks may be harder to manage than cyber risks, and urged Chinese companies releasing open-weight models to take safety seriously.

He acknowledged Anthropic's case for keeping advanced models proprietary but argued a mixed approach may work better, since open weights can still equip defenders. Tech360 reported that he pointed to Hugging Face reaching for a Chinese open model for cybersecurity after American closed models declined the task on guardrail grounds.

Tencent released its flagship Hy4 on November 4. The company says it excluded offensive hacking material from training data and kept defensive security content; internal reviewers reported strong basic biology knowledge but weaker performance at designing experiments. Weights and a safety-testing report are due in the coming days.

The talk lands amid a steady run of safety coverage on our tracker and alongside recent China AI stories we've been logging.