THE ESSENTIALS IN 30 SECONDS

  • Yao Shunyu, Tencent's chief AI scientist, said on Saturday in Hong Kong that model capabilities in biology will soon reach a level where real harm becomes possible, as in cybersecurity.
  • He said firms that release open models need to take safety very seriously, and suggested that open and closed models coexisting may be the more reasonable path.
  • Tencent plans to publish Hy4's weights and a safety testing report in the coming days, according to Reuters as cited by The Next Web.

Yao Shunyu, the chief AI scientist at Tencent, used a public appearance in Hong Kong on Saturday, November 7, to warn that AI systems are approaching a dangerous level of capability in biology. According to the South China Morning Post, as reported by The Next Web, he said models already pose risks in cybersecurity and that biology will soon follow, reaching a point where it is "actually possible to do harm." He added that biological risks may prove harder to manage than cyber ones. FourWeekMBA, drawing on reports by Yicai and IT Home, places the remarks at a dialogue held as part of the 2026 Future Science Prize events.

A message to open-model developers

Yao's warning was aimed in part at Chinese companies that publish open models, which anyone can download and modify. He said such firms need to treat safety very seriously. Asked whether Tencent would keep its Hunyuan models open if one became the world's top-ranked model, he gave no direct answer, according to Yicai as cited by FourWeekMBA. Instead he listed reasons labs open their models: attracting developers when they are not in the lead, sharing knowledge with the research community, and, increasingly, safety considerations.

Yao said US companies such as Anthropic make a strong case for keeping the most capable models closed, but argued that a mix of open and closed models may work best, since open models can also help defenders. He cautioned that attack and defence cannot be fully separated, because stronger general reasoning and coding can bring offensive skills even without attack material in the training data. He called for safeguards across the model's life cycle: careful data selection, safety reviews before release and monitoring afterwards.

Context: Hy4 and its safety report

The comments come days after Tencent launched its Hy4 flagship model on November 4. Tencent has said it left material on exploiting software flaws out of Hy4's training data and that internal expert reviews found the model weaker than leading systems at designing biology experiments; these are company claims that The Next Web says it has not tested. Tencent plans to release Hy4's weights and a safety testing report in the coming days, according to Reuters as cited by The Next Web.

Yao, 29, joined Tencent after about a year at OpenAI and was named chief AI scientist in December 2025.