Tencent AI chief warns models will soon pose biology risks, days after Hy4 launch
Tencent chief AI scientist Yao Shunyu said AI models' cyber abilities, and soon their biology abilities, will reach a threshold where harm is possible, according to the South China Morning Post. He spoke days after Tencent launched its flagship Hy4 model on 4 November.
Tencent's chief AI scientist has warned that AI models will soon be able to cause harm in biology, as they already can in cybersecurity. Yao Shunyu made the remarks on Saturday at a science event in Hong Kong, the South China Morning Post (SCMP) reported.
"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 said, according to the SCMP.
Open models and safety
Yao said Chinese companies that release open models, which anyone can download and change, need to take safety very seriously. He also warned that biology risks may be harder to manage than cyber risks.
He said US firms such as Anthropic make a strong case for keeping top models closed. However, he argued that a mix of open and closed models may work best, because open models can also help defenders.
Yao is 29. He joined Tencent last year after a year at OpenAI.
Context: the Hy4 launch
Yao spoke days after Tencent launched Hy4, its new flagship model, on 4 November. In its launch announcement, Tencent said it left material on exploiting software flaws out of Hy4's training data. The model was instead trained on defensive security data, the company said.
Tencent also said Hy4 has strong basic knowledge of biology but is weaker than leading models at designing experiments. The company said it based this on internal expert reviews. These are Tencent's claims, and TNW has not tested the model.
According to Reuters, Tencent plans to release Hy4's weights and a report on its safety testing in the coming days. The SCMP reported that benchmarking firm Artificial Analysis ranks Hy4 as the second-best Chinese model.
What this means
The warning comes from the scientist at the company that just released a flagship model, and it frames biology as the next risk area after cybersecurity. Yao's point that biology risks may be harder to manage than cyber risks is his own assessment, as is his view that open and closed models can coexist because open models help defenders.
Tencent's own statements on Hy4 line up with that framing: it says it excluded exploit material from training and rates the model as weaker than leading models at experiment design. Both claims rest on the company's internal expert reviews. The planned release of Hy4's weights and a safety testing report will be the first chance for outside parties to check them. Until then, the biology and security claims remain unverified by independent testing.
Related Articles
Tencent releases Hy4: 770B total parameters, 49B active, with native vision and computer use
Tencent has released Hy4, with 770B total parameters and 49B active parameters. It is the first Hy model with native vision and computer-use support. Weights are due to be open-sourced soon, and a lower-cost Hy4 flash is coming.
Tencent expands Hy4 inference cluster tenfold after WorkBuddy usage surge
Tencent says it has expanded the inference cluster serving Hy4 tenfold after many WorkBuddy users hit their quotas on the model's first day. Hy4 scores 35 on the Artificial Analysis Intelligence Index, second among Chinese models, and ranks 13th on Agent Arena.
Tencent's Hy4 adds vision and computer use, ranks first among open-weight models in blind finance evaluation
Tencent's Hy4 adds vision and computer use and is positioned in the top tier of open-weight models. On a WorkBuddy-focused evaluation it posts an average quality score of 83.18 and a 96.15% pass rate, and it ranked first among open-weight models in a blind finance evaluation by 213 experts.
Tencent releases Hy4 with 1.0M context and first visual understanding, priced at $0.417 / $1.251 per 1M tokens
Tencent has released the official version of Hy4, with a 1.0M-token context window and visual understanding for the first time. It is listed on OpenRouter at $0.417 input / $1.251 output per 1M tokens, shown as 50% off.
Comments
Loading...