AnswersPublished18 sources
Yao Shunyu’s AI Biology Warning—and What Tencent’s Hy4 Safety Results Show
Yao Shunyu warned that AI models may soon be capable of causing biological harm, as they already can in cyber. Hy4 launched on November 4 with visual understanding and computer use capabilities; Tencent said model weights and a system card would follow.
Published byEdited with GPT-6 LunaImages generated with GPT Image 2
Research answer
Key takeaways
- Yao Shunyu warned that AI models may soon be capable of causing biological harm, as they already can in cyber.
- Hy4 launched on November 4 with visual understanding and computer use capabilities; Tencent said model weights and a system card would follow.
- Tencent says it filtered some offensive cybersecurity material from training and had its security labs conduct red team audits.
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: What did Tencent’s 29-year-old chief AI scientist Yao Shunyu, who joined Tencent last year after a year at OpenAI, warn at the November 7 sc. Article summary: Yao’s central warning was that AI can already cause harm in cyberspace and is approaching a comparable threshold in biology, where separating offensive from defensive capabilities may be even harder. That is a warning ab. Topic tags: general, general web, government, academic, user generated. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, wate
Tencent chief AI scientist Yao Shunyu warned at a Hong Kong science symposium that AI models may soon be capable of causing harm in biology, much as they already can in cybersecurity. The warning points to a broader frontier-model risk; it does not establish that Tencent’s newly launched Hy4 has crossed a biological high-risk threshold. 2
8
What Yao warned about
Yao’s reported concern is that biological capabilities are approaching a point where they could enable real harm. Reporting also describes biological risks as harder to separate into offensive and defensive uses than cybersecurity risks. 1
8
The warning concerns the direction of AI capabilities, not a finding that every current model can carry out harmful biological work. Reports on Yao’s remarks also describe AI’s potential benefits in biology, including accelerating drug and treatment research. 8
What Tencent says about Hy4
Tencent launched Hy4 on November 4. The 770-billion-parameter model added visual understanding, reasoning and computer-use capabilities, including the ability to work through command-line and graphical interfaces. 4
5
At launch, Tencent described model weights and a system card with further safety information as planned releases. Those planned materials should be distinguished from the hosted launch of the model itself. 1
9
12
Tencent reported that Hy4 scored 30.4 out of 100 on Hy-BioExperimentBench, an evaluation of open-ended biological experiment design. The company said the score was below its high-risk threshold. Its internal assessment described Hy4 as having foundational biological knowledge but limited ability to turn that knowledge into actionable laboratory protocols. 4
13
That result provides context for Yao’s warning, but it is not a direct rebuttal: the warning concerns emerging capabilities across frontier AI, while the score is Tencent’s reported evaluation of one model. The benchmark result and the company’s threshold are not an independent guarantee that Hy4 is safe. 8
13
Tencent’s cybersecurity safeguards
Tencent says it limited cybersecurity training material to defensive content, such as proof-of-concept generation and vulnerability patching, and filtered out material including exploit development, reverse engineering and penetration testing. The company also says experts from Tencent’s Xuanwu Lab and Keen Security Lab conducted red-team audits. 12
These are safeguards Tencent reports using; they do not, by themselves, resolve the wider question raised by Yao: how to manage models as their scientific capabilities grow and potentially harmful uses become harder to separate from beneficial ones. 8
12
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "Yao Shunyu’s AI Biology Warning—and What Tencent’s Hy4 Safety Results Show"?
Yao Shunyu warned that AI models may soon be capable of causing biological harm, as they already can in cyber.
What are the key points to validate first?
Yao Shunyu warned that AI models may soon be capable of causing biological harm, as they already can in cyber. Hy4 launched on November 4 with visual understanding and computer use capabilities; Tencent said model weights and a system card would follow.
What should I do next in practice?
Tencent says it filtered some offensive cybersecurity material from training and had its security labs conduct red team audits.