Models
Tencent chief AI scientist Yao Shunyu says, per the South China Morning Post, that AI biology abilities will soon make harm possible and that Anthropic has a strong argument for keeping frontier models closed, while arguing that neither fully open nor fully closed may be ideal
Tencent’s chief AI scientist Yao Shunyu, per SCMP, says AI bio abilities will soon make harm possible and Anthropic has a strong argument for closed models.
Tencent’s chief AI scientist, Yao Shunyu, said at a science symposium in Hong Kong over the weekend of November 7–8, 2026, as reported by the South China Morning Post, that AI models’ cyber abilities and soon their biology abilities will reach a threshold where real harm becomes possible. He also said Anthropic has a strong argument for keeping frontier models closed, while arguing that neither fully open nor fully closed models may be ideal. Here is what that does and does not mean.
The essentials
- The warning: Yao said cyber abilities and soon biological abilities of AI models will reach a threshold where it is actually possible to do harm. He added that biology may be harder to regulate than cybersecurity because offensive and defensive uses are difficult to tell apart.
- On Anthropic: the SCMP headline, as relayed, says Anthropic has a “strong argument” for closed models. That is the SCMP’s characterization, and the article’s exact wording was not available to us. It is not an endorsement of Anthropic’s overall approach.
- His own position: Yao said neither completely open nor entirely closed models may be ideal, and that a mix could widen access while addressing security risks. He also noted open models can benefit defenders, and that AI can help smaller labs with research and drug discovery.
- Tencent’s own model: Tencent released Hy4 on November 4 and says its biological capabilities remain below the company’s own high-risk threshold, with a weight release planned. That threshold is Tencent’s internal measure, and we could not verify it independently.
What this is and is not
It is a senior researcher at a major Chinese tech company publicly acknowledging that the closed-model argument has merit. It is not a Tencent policy change, a decision to keep Hy4 closed, or evidence that any model can currently provide meaningful biological uplift. Yao’s statements are forward-looking (“soon”), and the quoted threshold language is as relayed by the SCMP and secondary outlets.
Why it matters
The open-versus-closed debate has mostly been framed as a US-versus-China argument. A prominent Chinese lab leader granting the safety case for closed models, while planning open weights himself, shows the debate is more mixed than the framing suggests. Watch whether Tencent’s Hy4 weight release includes extra safeguards or staged access.
What is not established
- A Tencent decision to keep any model closed: none reported.
- Evidence that current open models give meaningful biological uplift: not presented in the coverage we read.
- The SCMP article’s full text: not read; relayed by Techmeme and three other outlets.
- Independent verification of Tencent’s Hy4 risk threshold claim: none.
Summary table
| Point | What is stated as of November 7, 2026 | Source |
|---|---|---|
| Speaker | Yao Shunyu, Tencent chief AI scientist, ex-OpenAI | Madhyamam |
| Venue | Science symposium in Hong Kong, weekend of Nov. 7–8 | Madhyamam |
| On Anthropic | A strong argument for closed models (SCMP wording) | Techmeme (SCMP) |
| His view | Neither fully open nor fully closed may be ideal | Madhyamam |
| Hy4 | Released Nov. 4; bio capability below Tencent’s high-risk threshold | The Decoder |
FAQ
Did Tencent say it would keep models closed?
No. Yao argued a mix of open and closed approaches may be best, and Tencent plans a Hy4 weight release.
Did Yao endorse Anthropic?
He reportedly said Anthropic has a strong argument for closed models, which is narrower than an endorsement.
Are AI models already dangerous in biology?
Yao said capabilities are approaching a threshold; Tencent says Hy4 is below its high-risk level.
Method and limits
Sources checked on November 9, 2026: Techmeme’s summary of the SCMP report, Madhyamam, Archyde and The Decoder, read as search extracts. We did not read the SCMP article itself. Disclosure: Anthropic makes Claude, which helps draft articles on this site. Not legal or security advice.
Sources
- Techmeme (SCMP): Tencent's 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 (verified 2026-11-08)
- Madhyamam: Open AI models could enable bioweapon development, Tencent scientist warns (verified 2026-11-08)
- Archyde: Tencent AI Chief Yao Shunyu Says AI Models Could Pose Biological Risks (verified 2026-11-08)
- The Decoder: Tencent releases Hy4 with vision and computer use, claims second place among Chinese AI models (verified 2026-11-08)
Spotted an error in this article? Report it · Our corrections policy